sandraschi/freecad-mcp

View on GitHub ↗

FreeCAD 3D CAD automation via FastMCP — headless document/export plus FluidX3D/OpenFOAM CFD extensions. For Cursor / Claude Desktop. Webapp.

20 ★2 forksPythonUpdated 1mo ago

What you need to know

Exposes FreeCAD as a dual-interface server — an MCP server (SSE, port 10944) for AI agents plus a web dashboard (port 10945) for humans — covering STEP→STL conversion, BIM/Architecture, CFD (OpenFOAM + FluidX3D), FEM (CalculiX), 3D-printing slicing and model-marketplace search via 46 tools.

Install

just bootstrap && start.ps1

Usage

  • Connect any MCP client with url http://localhost:10944/sse (transport sse) and ask the agent to convert STEP files, design rooms, run fluid simulations or slice STL for printing.
  • Drive the same REST API (http://localhost:10944/api/v1/control/tool) directly from Python/curl for headless pipelines and parametric sweeps.
  • For fluid simulation choose OpenFOAM (CPU/Docker, turbulence, multiphase) or FluidX3D (GPU/OpenCL, fast uniform-grid iteration), or describe a flow problem in English and let an LLM write the config.
  • Run FEM with run_fem_analysis (CalculiX + Gmsh meshing, 10 material presets) and export point clouds to train surrogate models.

Key features

  • 46 MCP tools spanning geometry conversion, architecture/BIM (walls, slabs, columns, windows, doors, roofs, IFC import/export), FEM, CFD, slicing and marketplace search.
  • Persistent CAD file depot with full CRUD for managing parts and test brackets.
  • Fleet integration: QCAD DXF → FreeCAD → OpenFOAM/FluidX3D → Resonite/Godot/robotics handoff.
  • Natural-language CFD config (NL2FOAM) and PINN sampling data export for ML.
  • Tauri 2.0 + PyInstaller native desktop build option.

Best for

Engineers and makers who want an AI agent to run CAD conversion, simulation and slicing chains on FreeCAD, with a browser dashboard for humans on the same backend.

Caveats

  • Depends on heavy external tools: FreeCAD 1.1.1+, OpenFOAM 10 via Docker, FluidX3D via OpenCL (GPU), PrusaSlicer 2.8+, Ollama for local LLM.
  • Quick Start is Windows-oriented (start.ps1); ports are fixed at 10944/10945/10946.
  • MIT licensed.
Platforms: Local · WebClients: Claude Desktop · Cursor · Continue · Any MCP client

Reviewed 2026-08-18

Topics

cad-3dcfdcomputational-fluid-dynamicsfastmcp-3freecadheadlessmcpmcp-servermodel-context-protocolopenfoamwebapp
Stars
20★
Forks
2
Language
Python
License
Created
2026-05-07
Last push
2026-07-28