Changelog

[1.2.69] – 2026-11-11

Changed

• fdm_design says when and how a design agent researches before designing or asking: whenever the request names something that exists outside the person's hands (a product, a vehicle and year, a case, a standard part, a mounting place) or a decision turns on a fact about the world; six short steps (list the deciding facts, search each by exact name and generation, read the page not the snippet, rephrase before writing "not found", keep each value with its source, design around what stays unknown) sized to take a few minutes with any model that has web tools; and never to ask the person for what a search answers. Prompt text only; checked with the design-guidance tests; no printer contacted.

[1.2.68] – 2026-11-11

Fixed

• The fit guidance (fdm_fit, guidance://fdm/fit-by-function) derives a pocket's corners by offsetting the object's outline by the clearance. It used to say "at least the object's own radius", and a rounder pocket corner cuts into the object's corner even when the flats clear. Prompt text only; checked with the design-guidance tests; no printer contacted.

[1.2.67] – 2026-11-11

Added

• Serves two new design prompts and strengthens the everyday guidance. fdm_fit (also the resource guidance://fdm/fit-by-function) tells a design agent to choose each fit by what the interface has to do, never one clearance for every interface, keeping the object's size and uncertainty, the print correction, and the fit apart, with cited starting clearances and mechanisms that absorb uncertainty. fdm_flexures covers parts that flex, grip, snap, clip, spring, or hold something: compliant architectures, strain chosen by duty (once, daily, held bent), material and layer direction as planning bands, bite at the forced opening, heat and creep, and ease of use as a check equal to the hold (where the fingers go, the force to open, one-handed loading). fdm_design now carries the same judgement in short form for every task, including looking up published dimensions before asking, treating an unedited prefilled size as no measurement, designing to a cased product's size rather than the bare product's, fitting an object in its stiffest and largest state, treating printed screws at hand sizes as an ordinary choice with threads kept out of the main load path, and stating "prints without supports", "holds", or "fits" only with the engine check or slice number. fdm_assembly, fdm_after_slice and cad_design gain the matching fastener, review and parameter rules. The server's instructions say when to add each prompt. Supersedes the fit-by-function resource drafts by serving the fit text as a prompt as well, so prompt-only design workers read it. Validated with the design-guidance and CAD protocol tests against mocked transports; no printer contacted and no physical fit or print claimed. Prompt text only; no tool, slicing, or print behavior changes.

[1.2.66] – 2026-11-10

Fixed

• Allow a full-size Bambu A1 to start a checked job after a stopped job reports FAILED, print_type: idle, zero print error, empty HMS and a named prior job. Port Rowan McMahon’s A1 recovery contribution; retain fresh identity/nozzle checks and rejection of incomplete reports, active jobs and other models. Validated with mocked status regressions; A1 firmware evidence comes from the contributor, with no new physical test.
• Update Axios to 1.21.1 and basic-ftp to 6.2.6, including their upstream redirect-header and directory-download fixes. Keep bambu-node pinned.

[1.2.65] – 2026-11-10

Fixed

• Bambu Studio no longer hangs until SLICER_TIMEOUT_MS on a Mac where it once crashed. OrcaSlicer and FULU's OrcaSlicer-bambulab launch the same way and get the same change. The CLI opens a hidden window to draw plate thumbnails, which starts AppKit. When the person's crash history (~/Library/Application Support/CrashReporter/BambuStudio_<hardware UUID>.plist) dates a crash inside the app's last window restoration, AppKit raises its modal "reopen windows?" alert. The slicer then waits at 0% CPU, and every slice waited with it; stopping it never answered the alert. On macOS each Bambu-compatible slice now runs with a fresh, empty CFFIXED_USER_HOME of its own under TEMP_DIR, removed afterwards, so AppKit finds no crash history. The CLI reads its profiles from its own bundle or from the files this server passes, and sets no data folder without --datadir (Bambu Studio 02.08.02.61 src/BambuStudio.cpp), so slices are unchanged. With Bambu Studio 02.08.02.61 on macOS 27.2, plate_1.gcode was byte-identical between current main launched in a fresh private home and this build, for the same model (57m 47s, 40.78 g). Real-home launches were not compared (they can raise the alert). That the private home changes nothing else rests on the source above and on what was observed: AppKit created only Library/Application Support/CrashReporter in it, while preferences (cfprefsd ignores CFFIXED_USER_HOME), HOME, TMPDIR and the saved window state stay the person's own. A home seeded with that crash history hung current main at 0% CPU in NSAlert runModal and sliced on this build in 1.6 s. OrcaSlicer's results were not compared: the installed OrcaSlicer 2.4.0-dev crashed on every slice, before and after. Linux and Windows launches are unchanged.
• npm test runs the suite through scripts/test-with-temp-guard.mjs again, as 1.2.61 introduced it, so a suite that leaves temporary directories behind fails. A merge of the version files in 1.2.62 put back the old test script, so 1.2.62 to 1.2.64 shipped without the guard in the test script. The published code was unaffected: the package doesn't run its test script, and the guard was never part of the package. Test infrastructure only; checked by running npm test through the guard.

Added

• On macOS, a Bambu-compatible slicer that uses no CPU and writes nothing for SLICER_STALL_SECONDS (default 20; 0 turns it off) is stopped and reported with the stable structuredContent.code slicer-waiting-on-dialog, instead of after SLICER_TIMEOUT_MS. The message says to open the slicer once, answer what it shows and slice again; nothing is sent to the printer. It carries no slicer_report, because it is no verdict on the model.

[1.2.64] – 2026-11-10

Added

• Report CAD installation estimates/measured bytes and coded refusals; queue runs within their own deadlines, retain protected final-result validation records, reuse content-hash input bundles, and reveal preview openings with depth/feature edges. Clarify scan-reference scope and measured-interface precedence. Validated through confined headless fixtures; no physical fit/print or app update.

• Add optional confined code-CAD parity with bambu-printer-mcp: cad_status/setup/run/check_fit, cad_design and cad/1, pinned CAD/mesh libraries, protected scan inputs and explicit millimetre/Z-up preparation, independent checks and six renders. CAD_ENABLED=0 hides the family; setup installs only on request and requires internal ownership-enabled macOS storage without re-signing. Validated with stubbed printer transports and headless CAD; no physical print or installed Counterpart update.

[1.2.63] – 2026-11-10

Added

• counterpart_setup_fields says for each printer kind whether camera_snapshot reads its camera with only the settings the fields write (cameraSnapshot), as bambu-printer-mcp does. It is false for every kind here. For Bambu, camera_snapshot on this server reads only a PRINTER_CAMERA_SNAPSHOT_URL, which no setup field writes; bambu-printer-mcp reads the printer's own camera. For the other kinds a webcam is an add-on found through the printer's own settings once the server runs (counterpart_printer_info reports it), not from these fields. A client such as Counterpart can then ask for a camera picture in its setup check only where the MCP says it can, instead of deciding by the printer type's name. Every kind declares it; a client reads it missing as an older server. Checked with the setup-fields tests; no printer contacted.

[1.2.62] – 2026-11-10

Fixed

• Report printOutcome in print_3mf, upload_gcode and process_and_print_stl errors (counterpart/1): not-sent before any print command, unknown after a possible dispatch, or rejected when the printer refuses it. Fresh MQTT report timeouts no longer make clients guess whether printing started. Request-local tracking covers direct MQTT, FULU and all non-Bambu upload/start boundaries without matching error wording. Verified with mocked transports; no physical print started.

[1.2.61] – 2026-11-10

Fixed

• Track checked print snapshots from allocation so server shutdown also removes in-flight copies whose operation has not returned. Verified with an exiting child process and mocked native shutdown; no printer contacted.

• Remove test fixture directories at file teardown even after failures or early returns, and run the full Node test suite in a private temporary directory that fails on leftovers. Regression checks cover clean, leaking and failing suites; printer transports are mocked and no printer is contacted.

• Remove the server-owned temporary output directory on exit while retaining caller-specified TEMP_DIR contents, matching the Bambu fork. Checked through local stdio and HTTP server shutdown tests.

• On Node 18, an HTTP (streamable-http) server with a connected client now finishes its SIGTERM/SIGINT shutdown. Node 18's close() waits for clients' keep-alive connections (Node 19 and later close idle ones itself), so the server never exited, and a supervisor's later SIGKILL skipped the exit cleanup and left its temporary folder behind. Shutdown now closes the remaining connections. Found by the suite's leftover check in Node 18 CI (five leaked server folders and two timed-out shutdown tests), and checked locally with Node 18.20.8 and 22.

[1.2.60] – 2026-11-08

Fixed

• Drain FULU events after a polling sleep crosses the delivery deadline so a queued printer reply is read before disconnecting. Validated with mocked hosts under CPU load; no physical printer was used.

[1.2.59] – 2026-11-08

Fixed

• Bambu inspection accepts OrcaSlicer 2.5 dual-nozzle singleton nozzle_volume_type only when the selected plate declares both physical extruders and matching variants. Conflicting aliases and incomplete selections still reject, and independent hardware/material temperature limits remain enforced. Verified with synthetic X2D, H2C and H2D archives; no printer contacted.
• print_3mf delivers queued upload progress, including the final 100%, before returning its result. Notification failures log to stderr and do not fail the print. Verified through the public MCP handler with mocked dispatch and delayed/failing notifications; no printer contacted.

[1.2.58] – 2026-11-07

Fixed

• A large Bambu print_3mf upload no longer fails with "Timeout (data socket)" when a P1S's SD card pauses mid-file. The upload now waits up to 120 s for the printer to take more data. FTP commands keep their 15 s limit, so a printer that has really stopped is still reported. With bambu-printer-mcp on Nov 6, an 18.4 MB plate stopped at 13,287,424 bytes for more than 15 s; this server used the same 15 s limit. Checked with a local FTP server that pauses for twice the command timeout partway through a 12 MB file; no printer contacted by this server's tests.

Added

• Bambu print_3mf reports its LAN upload as MCP progress (counterpart/1) when the call has a progress token: progress is bytes sent, total the file's bytes, and _meta["counterpart/stage"] is "upload", at most once per whole percent, as bambu-printer-mcp does. Checked with unit tests and a local FTP server.

[1.2.57] – 2026-11-06

Changed

• When the upload of a print_3mf job fails, the error carries code: "printer-upload-failed" (counterpart/1), as bambu-printer-mcp 1.1.74 does: the file didn't reach the printer and no print command was sent, so a client knows the print did not start. Counterpart had reported such a failure as "could not confirm a print start". This server's upload already streams the file from disk, so it doesn't have bambu-printer-mcp 1.1.73's single-write upload timeout. Checked with the dispatch tests; no printer contacted.

[1.2.56] – 2026-11-06

Added

• fusion_mcp_status reports activeDesign when Fusion is connected: whether a design is open, read from the Fusion360MCP add-in itself. It calls the add-in's read-only get_design_type, which reads the active document's Design product the way every design command does (_design(), addon/server/command_handler.py:310-332 and :3555-3563 at the pinned 86a3016a5a5b). Success means a design is open. The add-in's own error kind NO_ACTIVE_DESIGN (addon/server/hints.py:17-19, rendered as **get_design_type** ERROR (NO_ACTIVE_DESIGN) by src/fusion360_mcp/server.py:122) means none is: Fusion is on its Home tab, and the pinned add-in refuses every design command, fusion_mcp_new_design included, until one is open (faust issue #12). Any other answer leaves activeDesign out (unknown). Clients such as Counterpart read this field instead of matching Fusion's error wording. Checked with the bridge's fake Fusion MCP: an open design, no design, another error, and the add-in not answering. Not run against a live Fusion.

[1.2.55] – 2026-11-06

Added

• counterpart_setup_fields marks OctoPrint's address field input: "url": it takes a host name or address, or an http or https URL with a path (OctoPrint behind HTTPS or a reverse proxy at a sub-path, which PRINTER_HOST accepts since 1.2.51). A client accepts a URL there and writes it as typed; other kinds' host fields take a host name or address only. Checked with the setup-fields tests; no printer contacted.

[1.2.54] – 2026-11-06

Fixed

• list_3mf_plates no longer leaves out a painted filament when a part's mesh is too large to read. A mesh over 64 MiB, or past 256 MiB read in total, used to be taken as unpainted, so a filament painted on it could go unlisted. Such a plate now lists every filament the project declares and says filaments_complete: false, so a client offers a spare row rather than miss one that prints. Every other plate says filaments_complete: true. A sliced plate is unaffected, even when it shares the unread mesh with an unsliced plate, because Studio's own plate_N.json names its filaments. Checked with unit tests for the one-file and the total limit.

[1.2.53] – 2026-11-06

Fixed

• list_3mf_plates gives each plate its filaments: the project filaments that plate prints with, and only those, with each one's colour, material, filament id and the first part using it (counterpart/1). Clients asked people to choose spools for every filament any plate used: on the Kass 3D scan rig's Screws plate, which prints only filament 3, Counterpart also asked for filaments 1 and 2 from other plates. A sliced plate uses Studio's own plate_N.json filament_ids. An unsliced plate is worked out as Bambu Studio's CLI does (PartPlate::get_extruders_under_cli, PartPlate.cpp:1298-1420; ModelVolume::get_extruders, Model.cpp:3357-3412): each part's filament, its painted facets, including meshes a component keeps in the root model (a component without p:path, bbs_3mf.cpp:4131-4148) (decoded as TriangleSelector::deserialize reads them, TriangleSelector.cpp:2111-2139), feature filaments, support and raft filaments, layer ranges and the plate's tool changes. A mixed filament counts as the filaments it mixes (PartPlate.cpp:1437-1449). Feature filaments and support follow what the slice applies rather than Studio's plate list where the two differ: a part that names no extruder prints with the project's wall, sparse and solid infill filaments, a part's or object's feature filament counts whether or not separate_filaments_for_features is set (PrintObject.cpp:3105-3150; libslic3r never reads that key), and support and raft resolve one setting at a time, the object's own else the project's (PrintObject.cpp:3052-3063), with support filament 0 following the object (Print.cpp:490-501). Negative parts, support blockers and enforcers, copies that aren't printable (checked per copy, as each build item carries its own flag: bbs_3mf.cpp:4183-4185, PartPlate.cpp:1311-1322) and other plates don't count; a mesh too large to read is taken as unpainted rather than failing the listing. Checked against two projects written by Bambu Studio 02.08.02.61's CLI on a Mac (two plates, three filaments; sliced and unsliced give the same answer as Studio's own slice) and the scan rig's metadata.
• print_3mf sends a tray only for the filaments the plate prints and -1 for every other project filament, on the LAN and plug-in routes, as Bambu Studio does (SelectMachineDialog::get_ams_mapping_result, SelectMachine.cpp:1463-1500). A caller's project-level ams_mapping with trays for unused filaments was sent as it was. Checked with unit tests; no printer contacted.

[1.2.52] – 2026-11-06

These port bambu-printer-mcp 1.1.72's fixes from a read of Bambu Studio 02.08.02.61's source. The real Bambu Studio slices named below ran through bambu-printer-mcp's code, which is the same as this release's for these paths; this repo's tests use those Studio-written files.

Fixed

• pause_at_layers keeps a project's own colour changes and pauses. Bambu Studio saves them in Metadata/custom_gcode_per_layer.xml (02.08.02.61 bbs_3mf.cpp:184), not custom_gcode_per_print_z.xml, so they were never read. --load-custom-gcodes replaces the project's whole map (BambuStudio.cpp:4005-4008), so a requested pause on a designer's file dropped its colour changes and pauses. They are now merged in. Without a pause request nothing is passed: the CLI keeps the project's own entries itself. A pause on a project with more than one plate needs slice_plate, because the CLI carries custom G-code for one plate only. Checked on a Mac with Bambu Studio 02.08.02.61: a project saved with a colour change at 0.6 mm and a pause at 1.0 mm, sliced with a pause before layer 8, kept all three. Slicing only; nothing was printed.
• Print safety no longer refuses every Engineering Plate job as "contradictory bed type metadata". Studio writes eng_plate in plate_N.json and Engineering Plate in the project settings (PrintConfig.hpp:437-456, PrintConfig.cpp:437-445). Every Studio name and token for each of its five plates now reads as one plate. Checked with a real Bambu Studio 02.08.02.61 Engineering Plate slice, which the previous code refused.
• An AMS-HT after the first is addressed as its own unit. Tray 129 is {ams_id: 129, slot_id: 0}, as Bambu Studio maps it (DevFilaSystem.cpp:367-368, SelectMachine.cpp:1489-1497), instead of slot 1 of unit 128, which doesn't exist. The fresh-state check finds that unit's tray the same way. Checked with unit tests; no printer contacted.
• When a slice fails without a usable result.json, the exit code is explained for every code Bambu Studio defines (libslic3r/Utils.hpp:21-75, 50 codes; 20 before), for example -66 (filaments can't be assigned to the nozzles) and -68 (a filament the nozzle can't print). A test checks the table against Utils.hpp.

Changed

• bed_type accepts supertack_plate, which slices as Studio's own Supertack Plate bed type. Checked on a Mac with Bambu Studio 02.08.02.61: the slice's plate_1.json says supertack_plate, the project says Supertack Plate, and the start G-code heats the bed to the filament's SuperTack temperature (45 °C for PLA). Slicing only; nothing was printed. OrcaSlicer takes the same name but wasn't sliced with it.
• slice_stl's skip_objects takes identify_ids, which the CLI matches against each object copy (BambuStudio.cpp:6530-6545), not positions. list_3mf_plates gives each plate's instances: every object copy with its identify_id from model_settings.config.

[1.2.51] – 2026-11-06

Added

• OctoPrint: octoprint_request_api_key gets an API key through OctoPrint's Application Keys approval (the person clicks Allow in OctoPrint), keeping the request open between calls because OctoPrint drops one that isn't polled every 5 seconds. counterpart_setup_fields offers it on OctoPrint's key as obtain, so a client can show an "Approve in OctoPrint" button instead of asking for a pasted key.
• OctoPrint: connect_printer connects OctoPrint to its printer with OctoPrint's saved port and baud rate, and with power_on first switches the printer on through PSU Control. counterpart_print_progress reports octoprint.disconnected (and a switched-off printer) instead of a bare offline, and a refused print says the same.
• OctoPrint: PRINTER_HOST may be a URL, for an OctoPrint behind HTTPS or a reverse proxy at a sub-path. The camera's relative and loopback URLs resolve against it as OctoPrint's page does in a browser (/webcam/… from the web server's root, webcam/… under OctoPrint's sub-path).
• npm run test:octoprint-live: an opt-in check against a real OctoPrint whose only printer is the Virtual Printer (it refuses any other).

Fixed

• PrusaSlicer configs with binary G-code on (Prusa's MK4, MK4S, MINI, XL and CORE One profiles once it is enabled) wrote binary bytes into the .gcode file, which inspection then refused. slice_stl adds --no-binary-gcode for such a config, and a slice that still comes out binary stops with an explanation.
• OctoPrint file names with spaces or other reserved characters are encoded when a file is read or started.
• OctoPrint's file list is cut to each file's name, path, type, size, date, estimated print time and last print, with folders' children (#4: a 324-file library was 500 KB).
• OctoPrint's cancel, start and temperature calls answer 204 with no body; their results now say what was done instead of being empty.

Checked against OctoPrint 1.11.8 and the 2.1 development build (octoprint/octoprint:canary) in Docker with OctoPrint's Virtual Printer, directly and behind an nginx sub-path proxy, through this server over stdio. The approval was given the way a browser gives it (sign-in, approval page, decision with OctoPrint's CSRF token). PSU Control is checked against a fake OctoPrint only, and PrusaSlicer's binary output with PrusaSlicer 2.8.1's CLI on Linux. No physical printer was used.

[1.2.50] – 2026-11-06

Added

• counterpart_setup_fields marks the kind this server runs when a .env sets no PRINTER_TYPE with default: true (OctoPrint, the same fallback the server uses), so a client recognises such a .env without a printer table of its own. Checked with the setup-fields tests; no printer contacted.

[1.2.49] – 2026-11-06

Added

• list_3mf_plates lists a designer's 3MF plates as Bambu Studio shows them before slicing: plate_count, and for each plate its index, name, objects, whether it is already sliced and, optionally, its own preview from the file. A plain 3MF with no plate metadata, even one with a large mesh, is one plate. Clients show a kit's plates from here instead of reading Metadata/model_settings.config themselves (Counterpart PRD §8.0 item 9). This ports bambu-printer-mcp 1.1.69. Checked with unit tests, and on real Bambu Studio projects with 6, 12 and 2 plates.

Removed

• slice_stl's split_to_plates and capabilities.slicing.plates (1.2.47). A designer's 3MF already carries its plates, and nothing used the split.

Fixed

• A slice whose result.json says success but writes no printable file reports slicer_report.failed: true.

[1.2.48] – 2026-11-06

Fixed

• fdm_slicing no longer says a print's own request carries the support settings. print_3mf's own slicing of a file that isn't sliced yet takes no support settings and uses the process profile's. The prompt now says a print gets the supports it was sliced with: print the sliced file itself, or slice the print with slice_stl and the same enable_support and support_type. Checked with the prompt tests.

[1.2.47] – 2026-11-06

Added

• slice_stl split_to_plates: a model bigger than the bed is split across plates. On the Bambu-compatible route, an STL or OBJ that doesn't fit the printer's bed is split into its separate parts, each kept whole. Parts laid out within 20 mm of each other stay together as the designer grouped them, and a part inside another stays with it, where they fit. They are arranged on as many plates as needed, inside the machine preset's printable area, clear of its excluded areas and under its printable height. The result is a deterministic multi-plate project 3MF, bed_split.plates (index, name, objects) and a plain message, and plate 1 (or slice_plate) of that project is sliced. A part larger than the bed is an error that names the part and its size against the bed's. counterpart_printer_info reports capabilities.slicing.plates. This ports bambu-printer-mcp 1.1.67. Checked with unit tests and the stand-in slicer; the same module split a real 850 × 521 mm kit STL into 4 plates, each of which Bambu Studio 02.09 on macOS sliced (through bambu-printer-mcp).

Changed

• A slicer's refusal reaches a person as a plain sentence. slicer_report.message and the start of a failed slice's text say what is wrong, for example Bambu Studio's and OrcaSlicer's empty-plate, off-plate and too-tall refusals, PrusaSlicer's "outside of the print volume" and CuraEngine's "no non-empty layers", matched on their own strings. The slicer's words stay in the new slicer_report.slicer_message for logs. A refusal without a result.json, including from PrusaSlicer and Cura, carries a slicer_report too. A successful slice's warnings are unchanged in message.

[1.2.46] – 2026-11-05

Added

• fusion_mcp_setup returns message when it isn't ready: what is missing (Fusion itself, the add-in at the pin, the port login item, or this server's Fusion settings, or a .env to write them to) and what to do, in plain words a client shows as they are. Counterpart's Mac app read the no-env-file outcome to word that itself. Checked with the setup tests.

[1.2.45] – 2026-11-05

Added

• blender_mcp_setup checks the MCP for Blender add-on that the configured bridge needs, when it is the mcp-for-blender (or blender-mcp) package run with uvx: the add-on file Blender loads (for the running Blender's version), its protocol and the one the package expects, whether Blender is listening and the running add-on's own get_addon_status. It changes nothing. With update: true it stages the add-on with the package's own install-addon at the inspected version, refuses an installed file that isn't a regular local MCP for Blender add-on or a staged one with another protocol, keeps the old file as a backup next to it and renames the new one into place. It never starts, stops or edits Blender. A handshake that can't be read is reported as live_error with the rest of the check. Overlapping updates run one after the other, and a cancelled request stops its installer and changes nothing more (in fusion_mcp_setup too). Its subprocesses (uvx and the package) get only the login environment and uv's and Blender's settings, never printer access codes or other server secrets. When the add-on can't be checked or updated here it says what to do in message. uvx.exe counts as uvx on Windows. The running Blender's version folder is found in Blender's macOS, Windows and Linux (~/.config/blender) layouts. Another bridge reports supported: false. Both tools declare remote scope read for a check and raw for update or install, so no paired device gains the change. Counterpart did this itself; it now comes from the MCP that runs the bridge, as in bambu-printer-mcp. Checked with a stand-in uvx and temporary add-on folders; the same code in bambu-printer-mcp checked the real package (2.2.7) and Blender 5.0 add-on on a Mac, without replacing it.
• fusion_mcp_setup checks, and with install: true installs on macOS, what the Fusion bridge needs: the Fusion360MCP add-in at the pinned commit (every file checked against its SHA-256; a symlink, an extra file or a changed byte installs nothing) set to run on startup, a login item that gives Fusion FUSION_MCP_PORT (installed only when launchd has it loaded), and this server's missing FUSION_MCP_* settings, added to its own .env after a copy of it. A FUSION_MCP_* setting the process has or the .env gives a value is never rewritten (an empty one, as .env.example leaves FUSION_MCP_COMMAND, is filled in), so a configured command or chosen arguments stay, and an add-in already at the pin is left in place. Without a .env it still installs the add-in and the login item and reports env: "no-env-file", with ready false until the settings are added. It reports whether Fusion is installed and never starts or quits it. Counterpart's Mac app did this itself. Checked with a synthetic pinned archive and a recording launchctl; the same code in bambu-printer-mcp reported a real Mac's setup ready and installed from the pinned GitHub archive into a temporary home folder.

Changed

• The Fusion bridge's default server (FUSION_MCP_PINNED_COMMIT) and the add-in fusion_mcp_setup installs are pinned to one commit, 86a3016a5a5b of DMontgomery40/fusion360-mcp-server: faust-machines' server and add-in with the document binding fusion_mcp_new_design's bind_document needs, while that change is reviewed upstream. Before, the default arguments named faust-machines 19b9f1e95192, while Counterpart installed the fork. This server's bridge doesn't use document binding; the fork has all 92 of faust's tool names at 19b9f1e95192 and adds new_document and close_document (compared from both commits' tools.py; schemas weren't compared). A server configured with only FUSION_MCP_COMMAND now runs the fork; one with its own FUSION_MCP_ARGS is unchanged.

[1.2.44] – 2026-11-05

Fixed

• get_slice_settings lists nozzle_diameter among the slice arguments it takes for own_process, so a schema-driven client checks layer heights against the same nozzle as the slice. It already read the argument; clients that send only listed arguments left it out, and a valid 0.42 mm layer for a 0.6 mm nozzle was refused. Checked with a stand-in slicer.

[1.2.43] – 2026-11-05

Fixed

• The FULU container test expected the real path on both sides of the config mount, so npm test failed on macOS, where /var is a link to /private/var. The server mounts the real folder at the path the host is given, and the test now expects that. Test only; found when bambu-printer-mcp's macOS CI ran the same test.

[1.2.42] – 2026-11-05

Added

• Every tool declares in _meta["counterpart/remote"] how a client that offers this server to other devices (Counterpart's remote endpoint) may expose it: scope, the least device scope that may call it (read, design, control, print, raw); raisedBy, a boolean argument that needs more (fusion_mcp_status launch, blender_mcp_status connect); and arguments, the names this server's configuration chooses (config, or config-top for the top-level printer type: the printer it targets, its credentials, hardware and programs) and the local files it reads (path) or writes (output). A tool that forwards an open object (an RPC payload, another server's tool arguments) declares every configuration name. A new tool or argument is classed here, with no Counterpart release. A script compared every listed tool's declaration with Counterpart's hand classification (main 1b66d937) and found no difference in scope or argument kind, so no paired device gains or loses a tool; the tests check every listed tool has one and hold the name floors. Checked by listing the tools over stdio; no printer.
• Every slice this server runs (Bambu Studio and OrcaSlicer CLIs, PrusaSlicer, Slic3r, Cura) keeps each finished slice, and slicing the same model the same way again restores it instead of running the slicer: Print Again, or printing a model a client just estimated, no longer waits for the slicer. A slice is identified by exactly what the slicer was given: its executable (path, size and modification time, so a slicer update slices again), this server's version, and its full argument list with every input file (model, profiles, filaments, generated overrides) replaced by its content hash, so editing any of them slices again. A run whose arguments name any other folder is never kept, nor is a failed slice; every check after the slicer still runs on a restored slice. The newest 16 are kept in ~/.cache/<package>/slices; SLICE_CACHE=off turns it off, SLICE_CACHE_DIR and SLICE_CACHE_ENTRIES move and size it. Counterpart kept its own copy of this keyed on the server's .env; it can now rely on the server. Presets the slicer finds by itself rather than through its arguments count only through its executable (a slicer update slices again). Checked with a stand-in slicer across a server restart; no real slicer.
• inspect_print_file returns sliced: the selected plate's estimate (seconds, grams, from the slicer's own slice_info.config) and settings (layer_height_mm, sparse_infill_percent, wall_loops, layer_count, from the 3MF's project settings and plate G-code, or a G-code file's own comments), and the file's plate_count. Slicing tools return slicer_report beside slicer_result: failed, message (the slicer's error and every plate warning, verbatim) and plate_count, on success and on a failed slice. Clients such as Counterpart read these instead of parsing slicer files themselves. Checked with synthetic 3MF and G-code files and recorded Bambu Studio result.json files.

[1.2.41] – 2026-11-05

Changed

• get_slice_settings' own_process answers for the slice a client would make. With source_path, the call also takes that slice's slicer_profile, template_name (its template), template_3mf_path, layer_height, wall_loops, sparse_infill_density, enable_support, support_type, quality and strength, as slice_stl takes them, with slicer_type and bambu_model as before. own_process is true only when that slice prints the process the 3MF carries, unchanged: a named profile or template replaces it, and an explicit value or a quality or strength other than standard changes it. Clients then pass their own configuration through instead of deciding which settings replace a designer's process. Checked with a stand-in slicer.

[1.2.40] – 2026-11-05

Fixed

• A relative BAMBU_NETWORK_CONFIG_DIR, per-call config folder or BAMBU_NETWORK_CERT_DIR is made absolute before it is mounted into FULU's container or sent to the host. The host runs from /plugins in the container, so a relative path pointed it at a folder that wasn't mounted. A certificate folder outside the config folder is created before the mount, so Docker doesn't create it as root. Checked with a stand-in docker running the fake FULU host.

[1.2.39] – 2026-11-05

Fixed

• fdm_slicing no longer tells a client that PrusaSlicer, Slic3r or Cura generate no supports. They take no support settings from slice_stl here, but a profile with supports enabled still generates them, so the text says supports follow the configured profile and to slice a checkpoint before modeling any. Checked with the prompt tests.
• Stopping the FULU bridge can no longer hang. With BAMBU_NETWORK_BRIDGE_IMAGE, the container is stopped (docker stop --time 2) and removed before the server waits for the docker client. The wait for the bridge's shell is bounded, so a descendant that keeps its pipes open no longer blocks shutdown or a restart. Checked with a stand-in docker and with a shell whose background child holds the pipes: 30 s before the fix (until that child exited), 4 s after.

[1.2.38] – 2026-11-05

Fixed

• Every extraction of a 3MF's process settings for slicing gets its own temporary directory, removed when the tool call that made it ends. 1.2.33 named them by content, so two slices of the same project at once still wrote one file, and a write could truncate the file the other slice was reading; nothing removed them either. The path get_slice_settings returns stays, named by content and written whole before it is renamed into place. Checked with concurrent extraction tests and a stand-in slicer.

Changed

• get_slice_settings on a 3MF with no slicer settings returns raw_key_count 0 (and no extracted_settings_path) instead of an error, so a client can tell a file with no settings to keep from one that couldn't be read. A file that isn't a readable 3MF, or whose settings file is there but can't be parsed, is still an error. A PrusaSlicer project's Metadata/Slic3r_PE.config is now read like bambu-printer-mcp reads it, so its settings are reported instead of refused. Counterpart refuses an edit of a designer's 3MF when it can't read the designer's settings, and needs this to allow edits of a 3MF that has none.

[1.2.37] – 2026-11-05

Added

• counterpart_discover_printers finds printers on the local network: Bambu Lab printers from their own SSDP announcements (serial, bambu_model value, name, firmware), and OctoPrint, Klipper (Moonraker) and PrusaLink hosts from their Bonjour adverts. It never connects to a printer; a Bambu port another app holds is network: "unavailable", not "none found". Checked with captured SSDP packets and dns-sd/avahi-browse lines; not run against live printers.
• counterpart_setup_fields lists every PRINTER_TYPE this server drives with the .env settings it reads for each (address, port, API key or password, and for Bambu the serial, model and access code), the Bambu model choices and the model each serial prefix names. Clients such as Counterpart offer printer setup from it instead of copying this server's tables. It also reports the slicer this server would run (slicer: type, path, whether it is there, and the Bambu profiles folder) with the SLICER_TYPE, SLICER_PATH and BAMBU_PROFILES_ROOT fields, required only when what they default to is missing, so a client can say "Bambu Studio isn't installed" or ask where it is instead of the person editing paths by hand.
• counterpart_propose_filaments proposes which loaded spool prints each of a model's colours before slicing, with an explicit verdict and note per part: a colour that isn't loaded says so instead of quietly printing another colour, and a colour that is loaded but not chosen (on the other feed, in another material, or on a spool another colour already prints from) says which. It checks a person's chosen spools against what is loaded now and maps painted slots; a sliced file's painted materials count when it chooses the feed. Each spool carries the filament profile slice_stl takes. An already-sliced file binds each filament to a nozzle, which the proposal can't read yet: on a dual-nozzle printer with both external spools, a mapping onto them is reported as external-nozzle-unverified instead of offered. Checked with mocked Bambu reports and fixture profile trees; not printed.

Changed

• get_printer_filaments slots carry label ("A2", "AMS HT A", "External", "Right external", "Spool") and filamentId, and a Bambu slot's name is the spool as a person names it ("Sunflower Yellow PLA Basic", from Bambu Studio's filaments_color_codes.json) instead of the raw "PLA Basic PLA #FEC600". An AMS-HT slot's id is its unit id ("128"), the tray value print_3mf takes, instead of ams_id × 4 + tray ("512").

[1.2.36] – 2026-11-05

Added

• fdm_slicing, a design prompt next to fdm_design: how this server's configured slicer orients a model and generates supports, and how to check a choice with a checkpoint slice_stl (read slicer_result; set orient, and always set enable_support, since leaving it out applies the preset's own setting). The text follows SLICER_TYPE and the same rules as capabilities.slicing: the Bambu-compatible CLI route orients and takes supports; generic OrcaSlicer takes supports but doesn't orient; PrusaSlicer, Cura and Slic3r take neither here. Counterpart wrote Bambu Studio's behaviour into its own design prompt; it now comes from the MCP that knows its slicer. Checked with the server's prompt tests.
• BAMBU_NETWORK_BRIDGE_IMAGE runs FULU's Linux host in a Docker image, as in bambu-printer-mcp, with BAMBU_NETWORK_BRIDGE_DOCKER, BAMBU_NETWORK_BRIDGE_DOCKER_HOST and an optional Colima profile that is started, never restarted, when Docker doesn't answer. It comes after an explicit bridge command and before a discovered FULU install. A Colima profile without BAMBU_NETWORK_BRIDGE_DOCKER_HOST uses that profile's own socket. Concurrent calls on a cold bridge start one container. The container runs with --init and a name of its own, and is removed when the bridge stops, so stopping the server no longer leaves FULU's host running. Checked with a stand-in docker and colima, and on a Mac against the real FULU host image in its own container: the agent started and wrote its logs and certificates to this server's folder, and no container was left afterwards. No print was sent through it.

Fixed

• FULU through this server no longer aborts on its config folder. The hand-written launcher it was pointed at mounted bambu-printer-mcp's agent folder, not this server's ~/.config/mcp-3d-printer-server/bambu-network, so the host couldn't open its log_dir (FULU #54). With BAMBU_NETWORK_BRIDGE_IMAGE the server mounts its own config folder and writes each print's checked snapshot inside it (Colima shares the home folder, not /tmp or /var/folders). A config folder reached through a symlink is mounted at the path the host is given. Snapshots a killed server left behind are removed after a day. Checked with a stand-in docker and colima running the fake FULU host, and against the real host image on a Mac (agent setup only; nothing printed).

[1.2.35] – 2026-11-05

Added

• get_slice_settings on a 3MF returns own_process: true when slice_stl, given the same slicer_type and bambu_model (both now accepted here) and no slicer_profile or template, slices that file with its own process (a designer's Bambu Studio or Orca project). It is the same test slice_stl uses, so a client can show that a designer's file prints with the designer's settings, and leave quality and strength at standard, without reading the 3MF itself. Counterpart's print sheet carried the last print's quality and strength over to a copy of a designer's 3MF, where they changed its layer height, walls or infill. Checked with a stand-in slicer.

Fixed

• 1.2.34's notes omitted one change it shipped: the preview leaves out filament-change flushing only between a material-switch M620 S… and its M621 S…, so setup forms such as the H2C start template's M620 O1 don't hide the part.

[1.2.34] – 2026-11-05

Added

• inspect_print_file takes include_preview: it then also returns preview, two PNG pictures drawn from the exact file's G-code, with that file's artifact_sha256. Every extruding line on the selected plate is drawn in its filament's colour at its extrusion width, including arcs (I/J or R, full circles too), without the start-up purge and prime lines or Bambu's filament-change flushing (M620 to M621). Features are read from Bambu Studio/Orca ; FEATURE: and PrusaSlicer/Cura ;TYPE: comments; G-code naming none is drawn whole. The first view is the top as it sits on the bed; the second is the bed side, turned over toward you so text on that face reads as it will. It never uses the plate thumbnails a sliced file carries: Bambu Studio's CLI copies a project's old plate PNGs into the sliced file unchanged, so they can show a part that isn't the one printing. Off by default, because the pictures are tens of KB of base64; it is meant for apps that show the person what will print, such as Counterpart's print sheet. On a designer's caliper card sliced on Nov 4, the preview shows the ruler ticks and vernier scale missing from the G-code, which is the defect that print had. Checked with synthetic G-code and with real Bambu Studio 2.8.2 sliced files.

[1.2.33] – 2026-11-05

Fixed

• A template's or project's process settings extract to a file named by their content, so two projects sliced at once can't overwrite each other's settings and slice with the wrong layer height, walls or infill. Checked with a concurrent extraction test.
• A 3MF that carries its own process settings (a designer's Bambu Studio or Orca project) now slices with them on the Bambu-compatible CLI route: slice_stl takes its layer height, walls, infill and the rest of its process unless the call names slicer_profile, template_name or template_3mf_path, and print_3mf's auto-slice does the same unless the call names slicer_profile. A configured default profile or template used to replace them. The machine still comes from bambu_model and the nozzle, so a project made for another printer can't bring its machine settings along. A 3MF with no process of its own, such as one from counterpart_make_3mf, still slices with the preset. Found when a designer's fold-and-snap caliper card (0.16 mm layers, 100% infill) was printed at the preset's values and no longer folded. Checked with a stand-in slicer. On Bambu Studio 2.8.2, that designer's file sliced through this same route (its own file as template_3mf_path) kept 0.16 mm, 100% infill, 2 walls and 6 top and 4 bottom layers, where the default route gave the preset's 0.2 mm and 15%.
• strength strong and max only raise walls and infill: a process that already has more walls or denser infill keeps them. Strong used to write 4 walls and 25% infill as given, which lowered that caliper card's 100% infill. On PrusaSlicer and Slic3r the floor is read against the loaded config's own perimeters and fill_density. An explicit wall_loops or sparse_infill_density is still written as given. Checked with a stand-in slicer only.

[1.2.32] – 2026-11-04

Added

• counterpart_printer_info reports capabilities.slicing: whether slice_stl can apply supports, copies (repetitions) and auto-orient with the slicer this server uses for that printer. The Bambu-compatible CLI route takes all three, generic OrcaSlicer takes supports, and PrusaSlicer, Slic3r and Cura take none. For a Bambu printer on OrcaSlicer, that assumes the call passes the printer's bambu_model. Clients such as Counterpart offer those choices from this instead of knowing which slicers accept what. Same field as bambu-printer-mcp 1.1.58. Checked with mocked transports only.

Changed

• slice_stl with repetitions above 1 on a slicer that can't make copies (anything but the Bambu-compatible CLI route) now says so and slices nothing; it used to slice one copy without a word. Checked with stand-in slicers only.

[1.2.31] – 2026-11-04

Added

• A printer refusal on print_3mf carries a stable code in its structured error: printer-command-rejected when the printer refuses the print command itself (Bambu HMS 0500-0500-0001-0007, MQTT command verification; the file is on the printer, nothing is printing), and printer-not-ready when the printer isn't in a state that can start a print. The message and suggestion are unchanged. Clients such as Counterpart can act on the code instead of matching the wording, which changes with printers and firmware. Checked with mocked MQTT only. Same change as bambu-printer-mcp 1.1.57.

[1.2.30] – 2026-11-04

Added

• slice_stl and slice_with_template take quality (draft, standard, fine) and strength (standard, strong, max). The server turns them into the slicer's own settings: Draft and Fine set the layer height for the nozzle in use (0.28 and 0.12 mm on a 0.4 mm nozzle, scaled with it), Strong and Max set walls and infill (4 walls at 25%, 6 at 40%), and Standard leaves the resolved preset alone. An explicit layer_height, wall_loops or sparse_infill_density still wins. Clients such as Counterpart can now send the person's choice instead of carrying slicer numbers. Checked with stand-in slicers only. For PrusaSlicer, Slic3r and generic OrcaSlicer the layer height scales with nozzle_diameter (or NOZZLE_DIAMETER), which must match the profile's nozzle; Cura takes only Standard. Same change as bambu-printer-mcp 1.1.56.

[1.2.29] – 2026-11-04

Changed

• With no method given, a FULU print over Bambu Cloud goes the way Bambu Studio sends it to a cloud-bound printer it can reach on the LAN (PrintJob.cpp): start_local_print_with_record, which uploads to the printer over the LAN and records the task in the account, in the account's cloud session. When that fails before the print command (the plug-in's -2010 to -2140 preparation results, other than a missing or oversized file), the same checked job goes once through start_print, with the printer checked again first. Any other failure may have reached the printer and is not sent again. A printer that reports no storage goes straight to start_print, as in Studio. Same change as bambu-printer-mcp 1.1.55. On a P1S with firmware 01.10.00.00, start_print alone returns -3120, HTTP 403 "internal blocking", every time. Errors name -2120, -2130 and -3120. Checked with the stand-in host and intercepted transports only; not yet on a printer.

[1.2.28] – 2026-11-04

Added

• slice_stl takes enable_support, support_type (normal(auto), tree(auto), normal(manual), tree(manual)) and support_threshold_angle, written over the selected process profile like layer_height and wall_loops. Bambu Studio and OrcaSlicer only; PrusaSlicer and Cura refuse them. Tested with the fake Bambu Studio CLI install; bambu-printer-mcp 1.1.54 checks the same keys against the installed Bambu Studio 02.08.02.61. No printer was involved.

[1.2.27] – 2026-11-04

Added

• Every tool now carries the four MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) from one table, src/tool-annotations.ts, so clients can tell reads from actions without parsing descriptions. Before, 7 of 45 tools had partial annotations. Destructive means the tool may end or replace something that exists: cancel_print, upload_gcode (same file name), camera_snapshot with save_path, get_slice_settings (a same-named 3MF replaces the extracted settings file), save_template, counterpart_print_route with a route, and the pass-through calls fulu_bambu_network_rpc, blender_mcp_call and fusion_mcp_call. Starting a print and heating are marked additive, matching the no-confirmation decision. Annotations are hints for clients; every safety check still applies. A test lists the tools over stdio and fails for any tool without all four.

Fixed

• Resume a Bambu job that the printer paused itself. A P1S paused by Pause or by a pause_at_layers stop reports print_error 0300_8013 ("paused by the user" / "paused by the pause command in the file"), and Resume refused it as a print error, so a paused job could only be resumed at the printer. The notice is excused only for Resume, only while the printer reports PAUSE; any other print error still refuses, and starting a job still needs no error. Ported from bambu-printer-mcp.
• FULU's BambuNetwork agent now receives the client headers FULU's own GUI sends before net.start (Bambu Studio 02.05.02.51 on Linux, with a per-install device id kept in the config folder). Bambu's cloud issues the printer-control certificate and creates print tasks for this client; without the headers the agent was an unknown client. BAMBU_NETWORK_CLIENT_NAME and BAMBU_NETWORK_CLIENT_VERSION override the name and version. Ported from bambu-printer-mcp.
• Evidence: mocked Bambu MQTT/FTPS boundaries and the fake FULU host only; no printer was contacted.

[1.2.26] – 2026-11-04

Changed

• Start prints and positive heating without a server-side confirmation prompt, on every printer type. The client's own print action is the approval and the printer runs its own checks, so clients without MCP elicitation support now print and heat normally. PRINT_REQUIRE_CONFIRMATION and BAMBU_REQUIRE_CONFIRMATION are removed. File inspection, hardware and material ceilings, declared materials, fresh printer state (read, rechecked before upload and again before dispatch) and Bambu identity checks are unchanged. Matches bambu-printer-mcp.
• resume_print asks a person only when the printer cannot hand the file back for re-reading (PrusaLink, Repetier, Creality), where the job is matched by name only. OctoPrint, Klipper and Duet (DSF) re-read the file, compare its SHA-256 and resume without a prompt. PRINT_CONFIRMATION_TIMEOUT_MS now bounds that prompt and the Bambu printer-model prompt.
• Stop treating a finished job as a bed-clearance case: a printer that reports a finished or cancelled job (Moonraker complete or cancelled, PrusaLink FINISHED or STOPPED, Creality finished, Bambu FINISH) starts the next print like an idle one. Matches bambu-printer-mcp's bed-check removal (bambu-printer-mcp #90).
• Evidence: mocked Bambu MQTT/FTPS boundaries and loopback HTTP printer APIs only; no real printer was used.

Fixed

• Include the authenticated control reply module in the npm package. The route imported it but the explicit package allowlist omitted it, so a packed install exited before MCP initialization. The packed-install smoke test covers startup and discovery through all three entrypoints.

• Keep authenticated FULU Pause, Resume and Stop request IDs as strings in Studio's 20000–29999 window, with a shared counter across LAN, cloud and retries that wraps inside the window. A cloud-bound P1S 01.10 rejects larger IDs with 0502400A even when their signatures are valid. Preserve reply correlation and the delivery hold; remove guidance claiming controls require Studio/Handy or Developer Mode. Automated regressions cover both transports and wraparound; hardware validation is limited to the maintainer's P1S.

• An ams_mapping object (and the 3MF's own mapping) is ordered by project filament index, as bambu-printer-mcp orders it. Before, its tray values were sorted, so {"0": 3, "1": 0} sent tray 0 for filament 0.

• Bambu printers slice with Bambu Studio when SLICER_TYPE is not set, as bambu-printer-mcp does. The previous default, FULU OrcaSlicer-bambulab, crashes on every command-line slice in its only release (v1.0.0, macOS DMG and Linux builds; fixes in FULU #56 and #57 are not released), so slice_stl, estimates and auto-slicing failed for every Bambu user without a slicer setting. FULU's Orca stays fully supported with SLICER_TYPE=orcaslicer-bambulab.

• A relative TEMP_DIR (such as ./temp) is resolved against the server's working directory, as bambu-printer-mcp 1.1.44 does. Bambu Studio fails every 3MF export to a relative output folder ("Project export to temp/…/…_sliced.3mf failed").

• Accept Bambu Studio 2.9 layer-temperature changes (M104 M S<temperature> with an optional T selector) in the Bambu file inspector. Preserve hardware/material ceilings, tool validation, and rejection of numeric markers, other heaters and extra parameters. Reproduced using isolated Studio 02.09.01.70 output; no physical print was run. This ports the fix prepared in bambu-printer-mcp #97.

• Update basic-ftp to 6.2.5, including its 6.2.4 fix for directory listings containing inherited-property fact names. Offline listing reproduction fails with 6.2.3 and preserves the file name and size with 6.2.5; no printer was contacted.

• Resolve H2C first-layer temperature tables by the selected filament index and extruder variant instead of treating every variant row as a filament. Preserve unique bindings, header agreement and independent base/offset ceilings. Ports the correction reported by Steve Pritchard (sjpritchard); an unchanged Studio 2.9 H2C slice passes offline inspection after reproducing the failure.

• print_3mf with a plate_index prints, inspects and checks pauses on that plate on every route: FULU's BambuNetwork route used to inspect and dispatch plate 1, and a pre-sliced file's requested pauses were checked against plate 1. Tested with the fake FULU host; no printer was contacted.

Added

• slice_stl, slice_with_template and print_3mf accept pause_at_layers, matching bambu-printer-mcp. Bambu Studio and Bambu-compatible OrcaSlicer CLIs place the printer's own pause before each requested 1-based layer (2 or greater), using a first slice to find exact heights. Existing project pauses and custom G-code are preserved. Output must contain every requested pause; already sliced files missing one are refused before dispatch. Other slicers refuse the option rather than ignoring it. Evidence: fake slicer/MCP tests only; no physical print or real slicer was run in this port.

• Bambu AMS inventory and tray choice, ported from bambu-printer-mcp:

  • get_printer_filaments reads the AMS from the printer's MQTT report for PRINTER_TYPE=bambu instead of refusing. The result carries bambu-printer-mcp's fields unchanged (trays with absolute slot, loaded, tray_type, tray_color, tray_info_idx and the resolved Bambu Studio/Orca filament profile; summary; recommended; load_filaments_all), plus counterpart/1 slots for every loaded AMS tray and the external spool (254, and 254/255 on dual-nozzle printers) with source: "printer". Optional bambu_model and nozzle_diameter pick model-specific profiles. counterpart_printer_info reports filaments: "ams" for Bambu.
  • print_3mf accepts ams_slots: one tray per used filament in plate order, expanded to the project-level ams_mapping from the plate's filament_ids and the G-code header, as bambu-printer-mcp does. ams_mapping also accepts the project-level array. Tray values are checked (0–15, 128–254, and 255 on dual-nozzle printers), ams_slots must name exactly one tray per used filament, and the fresh-state check still compares each mapped tray's material with the file. use_ams: false maps every used filament to the external spool of its nozzle on the wire. Counterpart's printer handoff sends exactly these arguments.
  • Evidence: mocked MQTT reports and FTPS/MQTT dispatch only; no real printer was used.

• Bambu pause_print and resume_print, and counterpart_printer_info reports pause and resume for Bambu, ported from bambu-printer-mcp:

  • Pause and stop (cancel_print) are never gated. Stop also retires the job's resume record.
  • Resume keeps bambu-printer-mcp's checked-job safety: only the job this server inspected and started, while a fresh report shows it paused under its unique name, with the fresh-state check repeated. Its bytes are re-read over FTPS (the uploaded 3MF, else the cache/<name>_plate_<n>.gcode a P1S extracts from it) and their SHA-256 compared; a changed file retires the record. A file this server uploaded that cannot be re-read refuses the resume. Only a file Bambu's plug-in placed (FULU's BambuNetwork route) that cannot be re-read asks a person through MCP elicitation.
  • Every MQTT report retires a record whose job ended or was replaced, even when nobody polls.
  • Checked 3MF prints are started with their unique remote name as subtask_name, as bambu-printer-mcp does, so the printer reports back which job is paused.
  • Evidence: in-process FTPS/MQTT fakes only; no real printer was paused or resumed.

• Bambu connection routes, including FULU's authenticated BambuNetwork route, ported from bambu-printer-mcp, so a P1S on authorization firmware without Developer Mode prints and is controlled over the local network:

  • One route decision (resolveConnection) for print_3mf, cancel_print, pause_print and resume_print: connection_mode (lan_mqtt_ftps, the default, or bambu_network) and connection_type (lan or cloud; cloud when unset, as in bambu-printer-mcp), defaulting to BAMBU_DEFAULT_CONNECTION_MODE and BAMBU_DEFAULT_CONNECTION_TYPE. A job is controlled over the connection that printed it. bambu-printer-mcp's bambu_native (X2D) and bambu_connect modes are refused with a clear error; nothing is sent over another connection.
  • print_3mf with bambu_network runs the same gates as the LAN path (inspection of the exact snapshot bytes, the AMS mapping, fresh LAN MQTT state before the bridge starts and again immediately before the print frame), then opens the plug-in's own session: net.start_local_print for lan, net.start_print for cloud. It sends Bambu Studio's ams_mapping, ams_mapping2 and ams_mapping_info, the Send dialog's calibration options as Bambu Studio resolves them, and the unique checked name as the task name. A synchronous -4030 is retried once after the certificate exchange; other non-zero results report the plug-in's code and messages. The job is recorded for a checked resume.
  • Stop, Pause and Resume over bambu_network go through the plug-in's authenticated session, held open until the printer answers; a refusal reports the printer's err_code, a -4/-4030 send is retried after reinstalling the certificate (three tries), and every event and send is written to control-trace.log in the bridge's config folder without the access code.
  • New counterpart_print_route tool, same contract as bambu-printer-mcp's: reads the route (lan or cloud, with the resolved mode and type) and changes it in the running server and in the .env file in its working directory when one exists (savedTo).
  • The bridge host is BAMBU_NETWORK_BRIDGE_COMMAND (bambu-printer-mcp's name), else FULU_BAMBU_BRIDGE_COMMAND, else the host check_fulu_orca_setup discovers next to FULU's OrcaSlicer-bambulab. Its config folder defaults to ~/.config/mcp-3d-printer-server/bambu-network (BAMBU_NETWORK_CONFIG_DIR). fulu_bambu_network_rpc still refuses raw print methods.
  • Evidence: a fake FULU host process that reproduces the real v1.0.0 host and plug-in behaviour bambu-printer-mcp recorded on a P1S, plus patched printer state; no real FULU host, plug-in, printer or Bambu Cloud was used by this server.

• Serves design guidance for agents that design or edit models to print as MCP prompts: fdm_design, fdm_multicolor, fdm_assembly and fdm_after_slice, plus the resource guidance://fdm/capability-guidance for bridges and overhangs. The server's instructions point every client to them. The text is David Montgomery's FDM design prompt pack of 3 November 2026, unchanged; bambu-printer-mcp serves the same guidance. Tested over a real stdio MCP connection; whether an agent follows it is a model behavior, not tested here.

Documentation

• README, docs/FULU.md, docs/SETUP.md and .env.example describe Bambu AMS inventory and tray choice, Bambu pause, resume and stop, the connection routes (BAMBU_DEFAULT_CONNECTION_MODE, BAMBU_DEFAULT_CONNECTION_TYPE, counterpart_print_route) and printing and controlling through FULU's BambuNetwork route, including its evidence limits here (a fake FULU host only).

[1.2.25] – 2026-11-02

Fixed

• lay_flat rests a model on a face that is its real bottom. Before, it turned the face-normal group with the largest area to face down and moved the lowest point to Z=0, without checking that those faces were the lowest. The stove gap filler from Scan to Make (six T strips) printed upside down: its cap undersides' 23,838 mm² face down but sit 12 mm above the fin tips, so it stood on its fins.
  • A pose qualifies only when some face, turned down, lies on the rotated model's lowest plane, within 0.5% of its height and at most 0.2 mm. The candidate poses are the 64 largest face groups and the model's six axes, so a domed part can flip exactly.
  • Among qualifying poses, lay_flat picks the one with the least part and support volume per mm² of bed contact. Support is the overhang more than 45° from vertical, times its height above the bed. Ties go to the larger contact, then to the face that already points down.
  • When no pose qualifies, the model is unchanged. The text result is still the STL path, and the new structuredContent {path, laid_flat, message} says why.
  • The gap filler now comes out cap-down, fin tips 14.4 mm up. This is checked from the output vertices only. The same rule in bambu-printer-mcp was also checked by slicing with Bambu Studio 02.08.02.61 for a P1S, which gave no warning; that slice was not repeated here.
• slice_stl returns Bambu Studio's own checks on the Bambu-compatible CLI route, and OrcaSlicer's where it writes the same result.json. When the CLI writes result.json (Bambu Studio 02.08.02.61 writes it on macOS on success too), structuredContent.slicer_result carries it parsed and unchanged: error_string, return_code and each sliced_plates[].warning_message. Before, the run's output directory was deleted with result.json in it, and warnings such as "It seems object … has floating cantilever. Please re-orient the object or enable support generation." were lost.
  • The text result stays the sliced file's path.
  • A failed slice is an error result whose text starts with Studio's error_string (unless it reports success), then each plate's warning_message, then the exit code, slicer output and suggestion, with no Error: prefix. Its structuredContent carries slicer_result beside the existing slicer details.
  • A result.json reporting a nonzero return_code fails the slice even when the CLI exits 0.
  • The G-code slicers (PrusaSlicer, Cura, Slic3r, generic OrcaSlicer without a Bambu preset) write no result.json and return structuredContent {path} only.
  • Tested with real result.json files from Bambu Studio 02.08.02.61 slicing the Scan to Make stove gap filler for a P1S on macOS, replayed by a fake CLI. The optional installed-Studio test also checks slicer_result on macOS; CI skips it. OrcaSlicer's result.json follows the same libslic3r format but was not run here.

[1.2.24] – 2026-11-02

Fixed

• fusion_mcp_status reports the Fusion360MCP add-in as reachable. The pinned add-in answers a ping on its fast path with {"ok":true,"status":"pong"}, and the bridge accepted only {"pong":true}, so it reported connected: false with a running add-in. Verified against a live Fusion with the add-in at 19b9f1e95192 through bambu-printer-mcp's identical bridge, and with socket fixtures here.

[1.2.23] – 2026-11-02

Removed

• Remove bed-clearance checks. A finished, complete, cancelled or stopped job prints like an idle printer. There is no "remove the previous part and debris" confirmation or forced human prompt, and no "the build plate is clear" line in print confirmations. The printer detects an occupied bed itself during bed leveling. Print confirmations keep their spool and file checks, and PRINT_REQUIRE_CONFIRMATION=0 now skips every print prompt. Verified with the full suite (570 tests) using mocked printers; no physical print was run.

[1.2.22] – 2026-11-02

Fixed

• camera_snapshot tries both 127.0.0.1 and ::1 for a localhost snapshot URL. Node 18 resolves localhost to ::1 first with no IPv4 fallback, so a webcam that listens on IPv4 only was missed. Verified with the OctoPrint Classic Webcam test on Node 18.20.8 and the full suite on Node 22.

[1.2.21] – 2026-11-02

Added

• Add the counterpart/1 tools that bambu-printer-mcp also implements, so one client can watch and control either server: counterpart_printer_info (server, printer identity and detected capabilities), counterpart_print_progress (one state, job, progress, time, layer, temperature and problem shape per firmware), camera_snapshot (bambu-printer-mcp's arguments and result plus capturedAt), get_printer_filaments (one external slot, with Spoolman through Moonraker), pause_print and resume_print. Each declares an outputSchema and returns structuredContent plus the same JSON as text; their errors carry text only, because MCP clients validate structured content against the schema even on errors.

• Pause is never gated. Resume reheats and moves the printer, so it is gated like a print start. It resumes only the job this server instance inspected and started, after a fresh report that the same job is paused, and after a person confirms through MCP elicitation (unless the server sets PRINT_REQUIRE_CONFIRMATION=0). OctoPrint, Klipper and Duet re-download the file and compare its SHA-256 with the inspected bytes before asking; PrusaLink, Repetier and Creality have no download route here, so the job is matched by name only. The printer is read again after the confirmation. Cancelling the job, or a report that it ended, retires it, so a later job that reuses the name is never resumed. A stop or heater-off request cancels a pending resume. A job started elsewhere, or before a restart, is refused. OctoPrint drops a resume sent while it is still pausing, so transitional states are reported as job.stage and resume waits for a settled pause.

• upload_gcode and start_print results name the remote job in jobName (for start_print, the checked copy that actually starts), matching job.name in counterpart_print_progress.

• slice_stl and slice_with_template accept layer_height, wall_loops and sparse_infill_density, written over the selected profile: Orca and Bambu Studio process keys on the Bambu-compatible and generic OrcaSlicer paths, and --layer-height, --perimeters and --fill-density for PrusaSlicer and Slic3r. Cura refuses them. Profile files are never modified.

• get_printer_filaments slots carry profile, the installed slicer filament profile for the slot's material, as an absolute path slice_stl takes as load_filaments: from Spoolman's material when Spoolman reports one, otherwise the configured FILAMENT_PROFILE; omitted when nothing resolves. OrcaSlicer bundled, system and user presets, and PrusaSlicer/Slic3r filament/*.ini presets, are matched exactly on filament_type, preferring a profile whose compatible_printers lists the printer in SLICER_PROFILE. bambu-printer-mcp's resolver did not fit: it looks up Bambu profiles by AMS tray_info_idx, which other printers do not report.

• Behaviour change: PrusaSlicer and Slic3r now load a filament preset. filament_profile, load_filaments or FILAMENT_PROFILE must be one .ini preset and is passed as a second --load after slicer_profile; before, it was silently ignored. A .json or a list is refused with instructions. Generic slicers (PrusaSlicer, Slic3r, generic OrcaSlicer) now honour load_filaments as well as its alias filament_profile; before, only the Bambu CLI path read load_filaments. For generic OrcaSlicer, a filament chosen for the call replaces the |filament.json default in SLICER_PROFILE instead of becoming a second slot, and a filament preset that inherits installed presets is passed to --load-filaments with its chain merged in (the CLI does not resolve inherits, so a bundled preset would slice without its own temperatures). A preset whose parent is not installed is passed unchanged.

• PRINTER_CAMERA_SNAPSHOT_URL overrides the camera source for any printer, and REPETIER_SLUG selects the Repetier-Server printer for the new tools.

• Add an optional Autodesk Fusion bridge (src/fusion-mcp-bridge.ts) to faust-machines/fusion360-mcp-server pinned at 19b9f1e95192, configured with FUSION_MCP_COMMAND, FUSION_MCP_ARGS (defaulting to the pinned server in socket mode), FUSION_MCP_TIMEOUT_MS and FUSION_MCP_HOST. fusion_mcp_status reports configuration, the Fusion process, whether the add-in answers its ping within a 2 s hard deadline (Blender's addon is not mistaken for it), and the server's tools; launch: true opens Fusion on macOS with --env FUSION_MCP_PORT=<port> and waits up to 90 s. The add-in port defaults to 9877 (PRD §8.9) because Blender's MCP addon holds the faust default, 9876; the bridge passes it to the Fusion server too. fusion_mcp_call forwards any discovered tool with its full content and errors and never retries. fusion_mcp_new_design opens a parametric design through execute_code. fusion_mcp_export writes <name>.stl, <name>.f3d and <name>.parameters.json, verifies the STL and its units against the Fusion body, and never overwrites. Unlike the Blender bridge, the client stays connected for the life of the server. Only FUSION_* settings reach the Fusion process. Closing stdin now shuts the server down, so a host that disconnects without a signal does not leave it, or the Fusion child, running.

Evidence

• The pinned Fusion server started through the bridge with uvx and listed its 93 tools; Fusion itself was not installed, so no design was built or exported live. The bridge's tests use a stdio peer that reproduces that commit's recorded --mode mock output. At that commit no tool works while Fusion shows its home screen without an open document (faust-machines issue #12), so fusion_mcp_new_design explains how to open one.
• OctoPrint 1.11.2 and 1.8.7 (virtual printer) and Klipper with Moonraker 1.5.0 on simulavr, with Spoolman 0.27.1: info, progress, camera snapshot, filaments, a checked upload-and-print through MCP with the confirmation answered, pause, resume (re-read bytes and its own confirmation), cancel, and a refused resume after the cancel. jobName equalled the reported job.name on all three. These are simulators, not printers.
• PrusaLink, Duet (standalone rr_* and DSF), Repetier-Server and Creality follow their API documentation and are covered by recorded-shape mapping tests and loopback HTTP mocks only. Repetier's done, printTime, printedTimeComp and @pause are not in its API manual. Creality's Moonraker ports (7125, then 4408) come from community K1 setups. Duet uses rr_model flags d99v instead of d99f, because f omits job.file.
• The Bambu mapping is tested offline and matches bambu-printer-mcp's encodings. This server has no Bambu camera, pause or resume path.

1.2.20 — 2026-10-30

Fixed

• Keep already-horizontal faces eligible when lay_flat chooses the largest flat face. Thin plates now rest on their broad face at Z=0 instead of standing on an edge, including after repeated calls. When up- and down-facing areas tie exactly, as for a plate with a raised block or lettering, keep the face that already points down so a seated part is not flipped. Validate with generated STL geometry in horizontal, upside-down, edge-on, and tilted orientations and with a raised block in either face order; no printer or live Blender scene is used.

Tests

• Verify malformed HTTP JSON returns a structured 400 response while the active MCP session remains usable. Exercise loopback HTTP only; no real printer or live Blender scene is used.

1.2.19 — 2026-10-30

Fixed

• Restore Node.js 18 support removed in 1.2.18. The MCP SDK's streamable HTTP transport calls the global crypto, which Node.js 18 lacks and @hono/node-server 2.x no longer installs; the server now provides it from node:crypto when missing. The package again declares Node.js 18 or newer and CI tests 18 through 24. Fresh npm installs on Node.js 18 resolve @hono/node-server 1.19.17, as they did for 1.2.18 apart from an engine warning. That version is inside GHSA-rmxm-3fg6-px4f's range, which affects only serveStatic; this server does not use it. Node.js 20 or newer installs the patched 2.1.5, which the repository lock keeps, so Node.js 18 CI and clones run it outside its declared Node.js 20 range. A regression starts the HTTP and SSE entrypoints without the global; the full suite and packed-install test pass on Node.js 18.20.8 and 24. Offline tests only; no printer is contacted.

1.2.18 — 2026-10-30

Changed

• Require Node.js 20 or newer. Node.js 18 reached end-of-life in April 2025, and the patched @hono/node-server relies on the global Web Crypto API that Node.js 18 does not expose by default, so streamable HTTP fails there. CI now tests Node.js 20, 22 and 24.

Dependencies

• Refresh the installation lock to @hono/node-server 2.1.5 for GHSA-rmxm-3fg6-px4f using npm audit fix without forcing dependency ranges. The MCP SDK already accepts this version. The updated dependency declares Node 20 or newer; no patched 1.x release is available. This lock controls reproducible installs, not existing consumers’ locks. Validation uses offline tests and mocked transports; no printer is contacted.

1.2.17 — 2026-10-27

Fixed

• Accept the official OrcaSlicer M620.17 T<nozzle> S<target> L-1 preheat for an unused nozzle on X2D, H2D and H2D Pro. Keep the target bounded by the hardware limit and every declared material, restrict the physical nozzle to 0 or 1, and do not invent a filament/AMS position. Ported from bambu-printer-mcp 1.1.38 following ZombieLurker's H2D report. Reproduced and validated here with offline file-inspection regressions; no physical print was run.

Dependencies

• Update the MCP SDK to 1.32.2 and Axios to 1.21.0 within their supported major versions. Verify MCP startup and HTTP header/body handling alongside the adapter and package tests. Thank you to the MCP TypeScript SDK contributors and Axios contributors.

1.2.16 — 2026-10-26

Documentation

• Show the documentation site's social card directly below the README title and link it to the site.

1.2.15 — 2026-10-26

Documentation

• Link the upstream reports for FULU's Linux host: the missing printer.cer (FULU #52), the segfault on printer-connection calls before a certificate folder is set (FULU #53), the abort on a missing create_agent log folder (FULU #54), and the abort when its input closes after an agent exists (FULU #55). This server refuses the methods behind #52 and #53. fulu_bambu_network_rpc can reach #54 with a missing log_dir, and #55 after net.create_agent once every request is answered; that abort is the one 1.2.14's validation saw.
• Link the proposed fixes for FULU's CLI crash, FULU #56 and FULU #57. With both, a patched macOS arm64 build answers --help and slices a P1S cube; FULU v1.0.0 crashes on both.
• Credit OrcaSlicer's fix for the 2.4.2 PartPlate::check_outside CLI crash: OrcaSlicer #14415 by packerlschupfer, released in 2.5.0-alpha. The STL transform workaround stays while 2.4.2 is the latest stable release; FULU's CLI has the same crash until it takes that fix.

Tests

• Write the STL manipulator smoke test's outputs to a scratch folder that it removes. Every npm test run used to leave six output folders in the repository's test/ directory.

1.2.14 — 2026-10-25

Ported from bambu-printer-mcp 1.1.34. Its evidence came from real P1S runs and an x86-64 Linux FULU container; this repository's validation is noted in each entry.

Fixed

• Let H2D printers with stock nozzles print. Every print stopped with Printer nozzle 0 information is unreliable, abnormal or unknown because the HS01 nozzles report stat: 1 permanently. Bambu Studio reads bit 0 of stat as "information unreliable" and bits 1-2 as abnormal or unknown. Nozzles that report no identity chip (sn: "N/A") now pass bit 0, and their configured diameter and type are still compared with the job. The Bambu print prompt names them so a person confirms they match the installed hotends, and checks log them once per printer on stderr, so a headless setup still records them. Abnormal or unknown nozzles still block, and a chipped nozzle with unreliable information asks for a refresh on the printer. Reported by MorganFreeeman in bambu-printer-mcp #67 with idle and mid-print H2D firmware 01.04.00.00 reports. Validated here with that report in mocked-status regressions; no H2D hardware was available.
• Say "declined or cancelled" when a person or the MCP client declines a hardware confirmation, instead of advising a client with elicitation support; clients without elicitation still get that advice. Ordinary print and heating prompts name the opt-out that applies to the printer: PRINT_REQUIRE_CONFIRMATION=0, or also BAMBU_REQUIRE_CONFIRMATION=0 for Bambu printers. Bed-clearance confirmations say a person must accept them. Also from bambu-printer-mcp #67; validated with mocked Bambu and generic adapters.
• Stop get_printer_status on Bambu printers from reporting UNKNOWN when the printer's first MQTT push is a sparse hello (model and modules). The read returned as soon as any data had arrived; it now waits, within its existing 1.5 s bound, until the full status has merged. bambu-printer-mcp saw this twice on a P1S; validated here with a delayed-push regression.
• Stop slice_stl option combinations from crashing OrcaSlicer's CLI. OrcaSlicer 2.4.2 segfaults in PartPlate::check_outside, which needs the GUI, when scale or rotate_x/rotate_y moves an object below the bed while arranging, the default for STL input. For STL input with OrcaSlicer or FULU, the server applies scale and rotations to a copy of the mesh in Orca's own order (scale, then X, then Y, then Z) and omits those flags; for 3MF input with arrange: true it refuses that combination with an explanation. repetitions on an STL now uses Orca's --clone-objects, which cannot be combined with clone_objects. The repetitions, arrange and clone_objects descriptions now state the CLI's real limits. Bambu Studio keeps every flag. bambu-printer-mcp verified the crash and fix against OrcaSlicer 2.4.2 on Linux, including the 10 × 20 × 30 mm rotation-order check; this repository's regressions use a fake CLI that records the mesh it receives.
• Report how a Bambu-compatible slicer failed: Bambu Studio and OrcaSlicer failures now name the CLI's own reason, from the result.json it writes on Linux or from its exit code, next to the existing exit code or signal and output tails. Validated with a fake CLI.
• Make FULU BambuNetwork bridge probes work with FULU's v1.0.0 Linux host. The host exits when its input closes, dropping requests it has not answered, so every probe and RPC call returned no responses; the server now keeps the input open until each request is answered. On Linux it passes PJARCZAK_BAMBU_NETWORK_SO and PJARCZAK_BAMBU_SOURCE_SO inside PJARCZAK_BAMBU_PLUGIN_DIR unless set, because the host opens them from its working directory (FULU #51). The default Linux bridge command pipes the host through cat, because the host exits 101 on the socket stdout Node.js provides (FULU #50); probes and RPC results explain exit code 101, and the FULU guide documents a wrapper for custom commands. A handshake from a host that could not load the networking plug-in is no longer reported as ready, and file open failed names the missing setting. A host that stops answering times out with its stderr, how many requests went unanswered and any hints, and Linux setup checks report the library paths that reach the host, including ones you set. macOS bundle notes and wrapper commands appear only for macOS, and the wrapper command only when the wrapper exists. Linux host bugs were reported with diagnosis and upstream issues by Aaron Roberts (aaroberts) in bambu-printer-mcp #68. Validated with a fake host that exits on end of input, and in the maintainer's x86-64 Linux container (QEMU emulation) against FULU's real v1.0.0 host and networking plug-in 02.05.02.58: probes were ready through the default command and the wrapper, a directly launched host returned exit 101 with its explanation, and a missing plug-in folder returned the file open failed explanation. net.create_agent through fulu_bambu_network_rpc created an agent; the host then aborted while exiting, after answering. No printer was contacted. bambu-printer-mcp's bridge LAN printing, certificate, -4030 and job-name changes do not apply: this server refuses raw bridge print and printer-connection methods and prints over its direct MQTT/FTPS path.

Documentation

• Add the FULU guide's Linux host setup and an "Upstream bugs and workarounds" table naming FULU #50 and #51, the FULU runtime's missing printer.cer, the FULU v1.0.0 CLI crash, the OrcaSlicer 2.4.2 check_outside crash (related to the closed OrcaSlicer #13328), and the Bambu plug-in's -4030 design, with what this server does about each.

Dependencies

• Move the direct zod dependency to 4.6.7 for GHSA-mrhx-c7xc-m3w5, memory exhaustion when validating oversized constrained collections; every zod 3 release is affected. The MCP SDK and zod-to-json-schema both accept zod 4, and this server does not import zod itself. Update body-parser to 2.4.0 (GHSA-27r3-jxmh-p828), fast-uri to 3.1.9, follow-redirects to 1.16.2, hono to 4.13.14 and ip-address to 10.7.5 for their new advisories. npm audit now reports no vulnerabilities. The documentation site's VitePress development server still pulls vite 6.4.3 and esbuild with advisories that have no VitePress 1.x fix; they affect only vitepress dev, not the published static site.

1.2.13 — 2026-10-20

Dependencies

• Refresh the reproducible installation/test lock from MQTT.js 5.10.4 to 5.16.1, including its fix for reconnecting when the previous stream is already destroyed. Keep bambu-node pinned at 3.22.21 and preserve the server's checked MQTT wire format and rejection guard. Thank you to Henk Kraal (hkraal) and the MQTT.js maintainers.
• Update the documentation build's source-map-js to 1.2.2 for GHSA-68fv-2mgg-jv7q, which addresses indexed source-map offsets blocking the event loop. Thank you to 7rulnik. Refresh compatible Node 18 typings and MQTT's required transitive packages.

The root lock controls npm ci, including CI and reproducible local installs; npm does not ship that lock to consumers. Existing parent projects can retain older transitive MQTT versions and must refresh their own locks. This is not a new published MQTT minimum or a claim that every advisory is resolved. Remaining nodemon/chokidar/braces and VitePress/Vite/esbuild development-server advisories have no compatible automatic fix; no major upgrade or downgrade is forced. Printer safety behavior is unchanged and no physical printer operation is performed.

1.2.12 — 2026-10-13

Fixed

• Correct fresh H2C (31B) and H2D Pro (239) returned-serial identity without trusting configured identity or suppressing conflicts. Accept a cloud print-job identifier in model_id only when another recognized observed model or returned serial identifies the printer; unknown actual model fields still reject. Port reports from Adam (@adamkaplan, fork #47) and JaviOFC (fork #36).
• Accept Bambu Studio 02.08 P2S per-variant nozzle tables alongside the selected plate header, and the exact P2S M145 P0/P1 airduct forms. Preserve selected extruder/variant conflicts and independent thermal checks. Port Izzy Mansurov's (@sapoepsilon) code and regressions from fork #33, including the subsequent selection-validation corrections.
• Accept the official H2C startup forms only inside its single bounded executable startup block, with independent base/derived nozzle and bed checks and declared filament positions. Keep configuration layer_change_gcode keys from ending startup; actual layer markers and extrusion still close it. Reject unchecked G383 U targets outside the allowance. Thanks to Adam for fork #48 and the exact vendor template evidence.
• Accept the official X2D bare material-switch B flag and exact M620.22 I<declared filament> P1 runout-purge form while rejecting unsupported variants and preserving heater ceilings. Thanks to Travis Cashman (@travismcashan) for fork #44. These are file-inspection changes; no native X2D helper is added to this server.

Dependencies

• Update basic-ftp to 6.2.3 for GHSA-r4jh-29pc-gfqr: rejected control-character passwords/resume paths no longer cause an uncaught exception or leave the client unusable, and rejected PASS errors omit the password. Keep bambu-node pinned and preserve the parent's direct MQTT wire-shape and rejection guard.

Validation uses mocked transports, the pinned official H2C template, unsafe-file mutations, and clean package installations. No physical printer operation is performed or claimed.

1.2.11 — 2026-10-06

This release also includes the safety, slicing, Blender, and documentation changes prepared as 1.2.10 below. Version 1.2.10 was not published to npm.

Fixed

• Send Moonraker's configured API_KEY as X-Api-Key on every request, including multipart uploads and remote-file inspection downloads. Thanks to Jack Manning (@EastArctica) for #30 and Klipper 0.13.0 authentication evidence. Loopback tests cover authenticated and trusted-client configurations; no maintainer physical print was run.
• Correct fresh Bambu A1/A1 mini serial identity (039 / 030) and accept the official A1/A1 mini M109 H wait parameter without weakening independently checked S/R temperature targets. Port the fork's fixes reported by @Steavie (#39) and implemented by @nitpreet22 (#41). Covered by mocked identity and file-inspection regressions.
• Keep model parsing, slicing, and STL diagnostics on stderr so they cannot corrupt stdio MCP messages, following @travismcashan's report in the fork's #44. Subprocess regressions check the compiled parser and slicer.
• Make Bambu file metadata lookup read-only through the shared FTPS client; remove the obsolete bambu-js dependency that created directories during reads.

Dependencies and interoperability

• Update the MCP SDK to 1.32.1, basic-ftp to 6.2.2, JSZip to 3.10.2, dotenv within version 16, and proxy-addr to its patched release. The production npm audit reports no vulnerabilities; unresolved nodemon watcher advisories are development-only and remain deferred rather than forcing an unsafe downgrade.
• Verify discovery, invocation, and STL export with upstream mcp-for-blender 2.1.9 and the former blender-mcp 2.0.0 package in isolated Blender 5.0.1. The exported fixture has 12 triangles and measures 40 × 20 × 10 mm. This verifies application interoperability, not a user's live scene or a print.

Discovery

• Add the official MCP Registry ownership field and server metadata for the npm package. The listing documents printer backend configuration and marks API keys and Bambu LAN access codes as secrets.

1.2.10 — repository changes, not published to npm

Safety

• Port the bambu-printer-mcp print safety gate to every Bambu print and heat route (print_3mf, start_print, upload_gcode with print, process_and_print_stl, set_printer_temperature). Each route inspects a private snapshot of the exact selected plate or G-code (every S/R heater target and tool change), enforces per-model hardware and declared-material ceilings, and checks a fresh MQTT report of the printer's model, serial, nozzle, state, errors and loaded filament before, after the human confirmation and again before dispatch.
• Ask a human through MCP elicitation before every print start and positive heating command. PRINT_REQUIRE_CONFIRMATION=0 (all printers) or BAMBU_REQUIRE_CONFIRMATION=0 (Bambu only) opts out for headless clients; a printer reporting a finished job still asks. Clients without elicitation are refused with instructions.
• Validate temperatures as finite, non-negative numbers before connecting. Heater-off (0) and cancel are never gated and cancel pending checked prints.
• Start remote files only after downloading and inspecting them, then start a uniquely named checked copy (Bambu, OctoPrint, Moonraker, Duet). Repetier, Prusa and Creality refuse remote starts because no verified download route exists; upload the local G-code with print=true instead.
• Gate OctoPrint, Klipper/Moonraker, Duet, Repetier, Prusa and Creality with shared thermal checks: independent nozzle/bed/chamber ceilings (defaults 300/120/60 °C, server-only PRINTER_MAX_NOZZLE_TEMP, PRINTER_MAX_BED_TEMP, PRINTER_MAX_CHAMBER_TEMP), material ceilings, a declared material for positive nozzle heating (G-code filament_type or the new material argument), and adapter state checks that refuse printing, paused, errored, offline or unreadable printers. Klipper macro parameters that name a heater are checked; macro bodies on the printer cannot be inspected.
• Stop again when a cancel arrives while an OctoPrint or Repetier upload-and-print is still uploading, so the cancellation wins.
• Pass nozzle_type and bed_type through process_and_print_stl slicing.
• Refuse RepRapFirmware M144 bed standby/activation, which heats to stored temperatures the job does not show.
• Refuse RepRapFirmware M568 A1/A2 tool activation unless the job itself set that tool's standby/active target first; activation-only commands would heat to an uninspected printer-side value.
• Fix process_and_print_stl, which only logged a temperature mismatch and kept printing and whose check missed R targets. Expected extruder_temp/bed_temp must now equal the sliced job's highest S/R target or nothing is uploaded.
• Stop print_3mf from uploading the original project when auto-slicing fails or the selected plate has no G-code.
• Behavior changes: print_3mf compares bed_type (default BED_TYPE, else textured_plate) with the sliced plate's bed metadata, so a file sliced for another plate or without bed metadata is refused until bed_type matches. Moonraker complete/cancelled, PrusaLink FINISHED/STOPPED and Bambu FINISH count as a finished job, so the first print after a previous job always asks a human to confirm the bed is clear, even with confirmation opted out. Generic print uploads need a plain filename so the started file is exactly the uploaded one.
• Refuse raw FULU bridge print methods, printer messages and unknown methods in fulu_bambu_network_rpc; allow_mutating_method now covers only agent/session setup methods.
• Publish Bambu safety commands with string sequence ids and keep bambu-node's internal parser errors (H2-series serials, PAUSE to IDLE after a cancel) from terminating the server, without patching bambu-node.
• Fix OctoPrint nozzle targets (tool0 instead of extruder) and use the DSF raw-text /machine/code and PUT /machine/file routes for Duet.
• Give people 10 minutes to answer confirmation prompts (PRINT_CONFIRMATION_TIMEOUT_MS) instead of the MCP SDK's 60-second default, and report an unanswered prompt as a timeout. Previously a slow answer was reported as missing elicitation support, which pointed users at disabling the prompt.
• Report what a Bambu printer did after a print command instead of assuming success: watch fresh reports (BAMBU_DISPATCH_CHECK_MS, default 15 s) and return dispatch: "started" or "unconfirmed", or fail with an explanation when firmware 01.08.05+ refuses the command (HMS 0500-0500-0001-0007, needs LAN Only Mode and Developer Mode). Found on a real P1S, where the old code reported "sent successfully" while nothing printed.
• List Bambu files over FTPS directly (ported from bambu-printer-mcp): the previous bambu-js listing returned empty folders on every failure, including on a healthy P1S with 286 files.
• Explain credential rejections (Bambu MQTT "Not authorized", FTPS 530, HTTP 401/403) as a changed access code or API key instead of advising a retry, and mark safety refusals as not retryable.
• Evidence: unit tests and MCP-level tests with mocked MQTT/FTPS boundaries and loopback HTTP printer APIs cover every gate. One real P1S run (firmware 01.08.05 or later, LAN Only Mode off) exercised the Bambu path: fresh MQTT status and AMS reads, the nozzle-type gate correctly refusing a job sliced for a stainless nozzle on a hardened-steel printer, human confirmation, and the checked FTPS upload. The firmware then refused the start command (HMS 0500-0500-0001-0007), so nothing printed; that finding produced the dispatch check above, whose logic is unit-tested but not yet seen against the firmware. No printer was heated. The other backends are verified against mocked APIs only.

Slicing

• Resolve BambuStudio, OrcaSlicer, and FULU OrcaSlicer-bambulab CLI profiles before slicing. The exact model/nozzle machine preset must exist in the selected installation, inherits and include chains are flattened, and the model's cli_config.json entry is validated. A raw inherited preset such as P1S 0.4 is no longer passed to BambuStudio, which rejected it with exit 239 ("process not compatible with printer").
• Stop before running the slicer on missing presets, wrong nozzle sizes, missing parents or includes, cycles, malformed profiles, missing process or filament files, and machine/process lists in slicer_profile.
• Report slicer failures with the exit code or signal, timeout state, and stdout/stderr tails, plus slicer-specific suggestions instead of printer connectivity advice.
• Give every STL edit and slice its own output folder, so concurrent requests for same-named inputs cannot replace each other's result before it is used or printed, and write prepared slicer profiles atomically.
• Extend bases under the model (Z-up) instead of along Y; extend_stl_base and process_and_print_stl previously added the base to the model's side.
• Require a nonempty Metadata/plate_<n>.gcode in Bambu-compatible output. A checksum-only archive, missing output, or stale file from an earlier run is an error.
• Add nozzle_type (and BAMBU_NOZZLE_TYPE) so jobs can be sliced for the installed nozzle. Bambu machine presets assume the stock nozzle (stainless steel on P1S/P1P/A1), and the print gate compares the job with the printer's report, so a P1S with a hardened-steel nozzle could never print a CLI-sliced job.
• Keep OrcaSlicer's absolute-extrusion normalization when the machine preset owns the layer-change G-code, so isolating machine settings cannot pair absolute E with the machine's G92 E0 resets.
• Drop machine settings carried by process or filament files, including template 3MF project settings exported for another printer, so they cannot replace the selected preset's start G-code.
• Add bed_type, load_filaments, load_filament_ids, filament_colours, template selection, and the Bambu CLI placement and transform options to slice_stl. Slicing accepts p2s, h2s, and h2c when the installed slicer includes those presets. Generic OrcaSlicer keeps its machine;process|filament path unless the call passes bambu_model.
• Add slice_with_template, list_templates, save_template, and get_slice_settings for a local template registry (BAMBU_TEMPLATE_DIR).
• Add BAMBU_PROFILES_ROOT, BAMBU_SLICER_PROFILE_DIRS, BAMBU_TEMPLATE_DIR, and BAMBU_TEMPLATE_3MF_PATH configuration.

Blender MCP

• Add blender_mcp_export_stl, which exports named objects from the live Blender scene to a new, verified STL. It writes world-space evaluated geometry without changing the scene, selection, or mode, checks the export receipt and triangle count, never overwrites files, reports bounding-box dimensions measured from the written file, supports scale, and warns when geometry looks like it was modelled in metres. Blender MCP's own export_scene writes GLB/FBX only, so edits made through execute_blender_code previously had no verified path to a printable STL.
• Report measured bounding boxes for every verified STL, including blender_mcp_edit_model results.
• blender_mcp_status lists tool names and summaries by default (about 8 KB instead of 60 KB against mcp-for-blender 2.1.1) and returns full input schemas through tool_names or include_schemas.
• Surface the Blender addon's own exception message when an export fails, and point configuration errors at the renamed mcp-for-blender package and the addon's GUI-only requirement.
• Add scripts/blender-live-check.mjs, an opt-in check against a real, isolated factory-settings Blender. Verified with Blender 5.0.1 and mcp-for-blender 2.1.1, together with an end-to-end phone-case refit (import, measure, refit, fit check, export, and a real BambuStudio P1S slice). No physical print was made.

Documentation

• Restructure the README around a copy-and-paste "Set up with your agent" request and "What to ask your agent" examples, move manual configuration to docs/SETUP.md, and add docs/SLICING.md, docs/FULU.md, docs/BLENDER.md, and CONTRIBUTORS.md.
• Publish a generated documentation site to GitHub Pages (https://dmontgomery40.github.io/mcp-3D-printer-server/).
• Make AGENTS.md the tracked source of truth for release, review, community triage, safety, and Blender rules.

1.2.9

• Address Bambu FTPS upload failures reported in #22 by using TLS 1.2 control and data channels with explicit host identity and verified session reuse, including the Node 22/24 wrapped-socket session regression. Physical X1C firmware confirmation remains outstanding; no printer commands were sent in testing.
• Port standard Blender MCP discovery, tool forwarding, and verified STL edit support from the Bambu fork. Preserve this server's legacy shell/stdin bridge.
• Validate binary STL edits in bounded memory and cap ASCII inputs at 4 MiB.
• Apply edit deadlines and cancellation to STL validation as well as MCP calls.
• Require the Bambu model on raw print starts and upload-with-print, matching the fork's safety checks. Keep uploads without printing available.
• Isolate server and inline-upload scratch paths to prevent filename traversal and collisions from overwriting or deleting unrelated files.
• Refresh vulnerable dependencies, pin the supported bambu-node API version, declare the TypeScript build dependency, build before npm packing, and report the actual package version during MCP initialization.
• Keep generated dist files out of Git and validate clean builds, protocols, package installation, and Docker builds in CI.