Apps · Fusion 360 MCP
From CAD to catalog, in one prompt.
Watch it move left to right: a parametric part takes shape in Fusion 360, becomes a photoreal hero render in Blender, then lands as a print-ready A4 spec sheet. One prompt drives the whole chain over MCP — no engineer needed after you hit enter. End-to-end in about 28 seconds on a laptop.
How it's made · four steps
One agent, three apps, zero handoffs.
The orchestrator plans the build; Fusion 360, Blender, and the layout renderer are each a small set of schema-validated MCP tools. Swap the orchestrator — Azure AI Foundry, OpenAI, any tool-calling LLM — and the rest of the pipeline doesn't change.
Step 01
The model reads the prompt
“Spec sheet for the new heat sink.” It plans the build, picks the MCP tools, and drives the rest.
tool router
Step 02
Fusion 360 builds the part
User parameters, sketches, extrudes, one multi-edge fillet. Schema-validated tools — no freeform scripting against the CAD API.
set_user_parameter · extrude · export_stl
Step 03
Blender renders the hero
STL imported, brushed-aluminium shader, white-studio scene, Cycles on the GPU. Seconds, not minutes.
execute_blender_code · cycles
Step 04
The spec sheet composes
Hero render, dimensioned drawing, and a parameter table become an A4 PDF in your brand. Vector text, embedded images.
compose · A4 PDF
Inside the loop
Schema-validated tools, not freeform code.
Each MCP tool has a typed input schema and a typed return shape. The model can't hand garbage to Fusion's API — the bridge rejects it before a single call fires. That's the guardrail an LLM cannot bypass. The terminal below shows the actual tool calls behind the spec sheet.
9
user parameters drive the part
0
freeform code paths in CAD
28s
prompt → PDF on disk
100%
parametric · re-runnable
// one prompt, three apps, one PDF user> Generate the spec sheet for the new heat sink. 12 fins, 5.5 mm pitch, M4 mounting. ▸ fusion.set_user_parameter ×9 → 9 parameters ▸ fusion.create_sketch + extrude ×17 → 1 body · 82 cm³ ▸ fusion.fillet_all_edges { r:0.5mm } → 192 edges ▸ fusion.export_stl { refinement:high}→ heat-sink.stl · 3.1 MB ▸ blender.execute_blender_code { cycles, AgX } → hero.png · 2400×1500 · 14s ▸ spec-sheet.compose { template, rev:A } → spec_RevA.pdf · A4 assistant> Spec sheet ready. 28.3 s elapsed.
What's customisable
One pipeline, your brand and product family.
Brand & layout
Logo, colour, typography, layout grid — driven by a template you configure once. No code to change between brands.
Product family
Heat sinks today, hydraulic manifolds or cabinet hardware tomorrow. Each family gets a Fusion template and a parameter contract.
Render style
Studio white, shop floor, exploded view, ghosted internals. The Blender scene is just another input per template.
Spec-table schema
Material, finish, certification, packaging — you define which fields appear. Pulled from CAD parameters, ERP, or a YAML side-car.
Languages
Spec sheets in NL, FR, EN, DE from one source, with localised labels and units — handy across BeNeLux and DACH.
AI orchestrator
Claude today, Azure AI Foundry or OpenAI tomorrow. The MCP layer is provider-agnostic — any tool-calling LLM drives the same pipeline.
CAD support
Runs on any CAD with an MCP server.
Ready today · ~2 weeks
- Fusion 360 — Autodesk-announced official + community servers
- SolidWorks, Autodesk Inventor, Onshape
- Rhino + Grasshopper, FreeCAD, OpenSCAD, AutoCAD family
Custom bridge · 4–6 weeks
- Siemens NX — NXOpen API ready, no public MCP yet
- Dassault CATIA, PTC Creo
- Solid Edge, IronCAD
Most ‘ready today’ servers are community-maintained; Autodesk has announced official servers for Fusion, Revit and the AEC Data Model. Built on the open fusion-general MCP server; paired with blender-mcp for the hero render. The pipeline doesn't change between platforms — only the CAD-side bridge does.
Want this for your catalog?
If you've a Fusion 360 (or SolidWorks, or Inventor) library and a sales team copy-pasting spec sheets from old PDFs, book a call. We'll have a working pipeline against your parts in about two weeks.