Makes · Autodesk Fusion

Fusion MCP

Design in Autodesk Fusion by giving an AI design tasks instead of writing scripts.

My Fusion MCP is a local MCP server plus a Fusion add-in. An AI client like Codex builds and inspects native Fusion models through 25 tools, or runs one complete Fusion Python modeling script in a single batch. It’s how GPT Astra modeled Branca, my Ender 3 Pro extruder.

GitHub Get it on GitHub

v0.4.1 · 25 tools · PolyForm Noncommercial

How it connects

YouDesign intent
CodexPlan + operations
Fusion MCPTool connection
Autodesk FusionThe native CAD model
Model data + results back to Codex
  • Batch operations
  • Inspect the model
  • Save checkpoints

It started with one question

I asked Codex: can you build an MCP so I can design in Fusion by giving you design tasks instead of writing Python scripts?

The answer is this tool. GPT Astra used it to model Branca, a new extruder for my Ender 3 Pro, as a fully parametric Fusion design.

What it can do

Model

Sketch, extrude, fillet and joint tools, plus named user parameters, all on native Fusion geometry.

Batch

Run one complete Fusion Python modeling script as a durable job, with progress, a checkpoint before it runs and a model health check after.

Inspect

Nested assembly and geometry inspection: components, bodies, faces, edges, joints and parameters. It can also read what you select in Fusion.

Select

Pick edges and faces by geometry: type, direction, size or position, or by the feature that made them.

Check

Interference and clearance checks, minimum-distance measurement, motion sampling and feature and joint health checks.

Export

F3D, STEP and STL export, print-oriented copies and exact-size image capture.

How a build runs

One modeling script, tracked as a job from start to finish.

  1. Inspect. inspect_design finds the active design and its root component.
  2. Submit. run_model_script queues a complete modeling script as a durable job and returns right away. The full design is checkpointed before any code runs.
  3. Track. job_status and job_result report progress and structured results: created components, bodies, sketches and features, measurements, files and warnings. Resending the same operation ID returns the same job instead of running it twice.
  4. Check. Interference, clearance, motion samples and feature and joint health checks confirm the result.
  5. Export. F3D, STEP or STL, with a checked export that runs the modeling-hygiene audit first.

All 25 tools

Full signatures are in the tool reference on GitHub.

Rules and hygiene

  • get_modeling_rules
  • audit_modeling_hygiene

Design and sketch

  • new_design
  • inspect_design
  • rectangle_sketch
  • circle_sketch
  • extrude
  • set_parameter
  • fillet_edges
  • create_joint

Batch jobs

  • run_model_script
  • job_status
  • job_result
  • list_jobs
  • cancel_job

Inspect and select

  • inspect_assembly
  • inspect_full_assembly
  • inspect_body
  • read_selection
  • select_geometry
  • fusion_capabilities

Check

  • measure_minimum_distance
  • check_interference
  • validate_model

Export

  • export_model

Parametric by default

The MCP ships with my Fusion modeling rules and hands them to the AI, through a tool and an MCP resource, before any design task.

  • Component first. Create, name and activate the component before modeling in it.
  • Name immediately. Components, bodies, sketches and features are named by purpose.
  • Constrain sketches. Fully constrained, with an intentional anchor.
  • Drive real geometry. Named user parameters drive the dimensions that matter.
  • Healthy timeline. Warnings and errors are resolved at the source.
  • Preserve interfaces. Functional bores, datums and seats stay protected.
  • Evidence before release. Audit, exercise a driving parameter, then record it.

For purchased hardware, the rules go in order: find the real CAD online, ask me, rebuild it from the spec sheet, and only then use a placeholder. audit_modeling_hygiene reviews naming, sketch constraints, timeline health and parameters across the whole design.

Read the modeling rules →

Made with it

Branca, the Ender 3 Pro extruder

GPT Astra modeled Branca through this MCP: four printed parts built around the hardware already on the printer.

Branca extruder exploded view

What you need

Installation and configuration guides are in the repo.

  • Autodesk Fusion, installed and open (Windows is the documented install path)
  • Python 3.10+ for the MCP server
  • A local MCP client that runs STDIO servers, like Codex

Licensed under the PolyForm Noncommercial License 1.0.0: use, copy, modify and share it for any noncommercial purpose.

GitHub

Build something with it.

Point your AI client at Fusion and give it a design task.

Get it on GitHub

See what it built: Branca →

Branca R59 assembled