freecad-tool-server
Enables AI agents to programmatically build, analyze, and export 3D CAD geometry using FreeCAD through REST or MCP tools.
README
FreeCAD Tool Server
An HTTP + MCP tool server that lets an AI agent of your choice drive FreeCAD programmatically — building, analyzing, and exporting real 3D CAD geometry through typed tool calls.
Agents call REST endpoints (or MCP tools) to create primitives, run boolean operations, loft airfoils, drill holes, fillet edges, check clearances, run finite-element analysis via Elmer, and export validated STEP files — receiving structured feedback at every step.
Your agent (Claude Code, Claude Desktop, any MCP/HTTP client)
│ MCP tools or JSON over HTTP
▼
FreeCAD Tool Server (main.py, typed models in models.py)
│ in-process import
▼
FreeCAD 1.1 bundled Python ──► OpenCASCADE geometry kernel
The server runs inside FreeCAD's bundled Python interpreter — FreeCAD is imported as a
module, not shelled out as a subprocess. That's why you launch it with FreeCAD's python.exe,
not your system Python.
Prerequisites
- FreeCAD 1.1+ installed (ships its own Python 3.11).
- (Optional, for FEM) Elmer and
Gmsh on your
PATHfor the/fem/*endpoints andelmer_*.pyrunners.
No system Python is required to run the server — FreeCAD's interpreter provides everything.
Setup
1. Install the Python dependencies into FreeCAD's interpreter
Windows
"C:\Program Files\FreeCAD 1.1\bin\python.exe" -m pip install -r requirements.txt
Linux / macOS (adjust to your FreeCAD install)
/path/to/FreeCAD/bin/python -m pip install -r requirements.txt
2. Start the server
Windows — edit the FreeCAD path in start.bat if needed, then:
start.bat
Linux / macOS — set FREECAD_PYTHON to your FreeCAD interpreter, then:
FREECAD_PYTHON=/path/to/FreeCAD/bin/python ./start.sh
The server listens on http://localhost:8000. Interactive API docs (Swagger UI) are at http://localhost:8000/docs.
3. Health check
curl http://localhost:8000/health
# {"status":"ok","freecad_available":true,"freecad_version":"1.1.1",...}
Connect your agent
Option A — MCP (Claude Code, Claude Desktop, any MCP client)
mcp_server.py exposes the tool server over the Model Context Protocol. Point your MCP client
at it. For Claude Code, add it to your MCP config:
claude mcp add freecad -- "C:\Program Files\FreeCAD 1.1\bin\python.exe" /path/to/mcp_server.py
or add an entry to your client's MCP settings JSON:
{
"mcpServers": {
"freecad": {
"command": "C:\\Program Files\\FreeCAD 1.1\\bin\\python.exe",
"args": ["/absolute/path/to/mcp_server.py"]
}
}
}
The agent then gets every FreeCAD tool as a native MCP tool.
Option B — Claude Code reading AGENTS.md / CLAUDE.md directly
Clone this repo and open it in Claude Code (or any agent that reads AGENTS.md). The bundled
AGENTS.md is a complete operating manual — the modeling loop, every endpoint, the
common pitfalls, and the required integration checks. Start the server, tell your agent what to
build, and it drives the HTTP API using that guide. A copy is provided as CLAUDE.md so Claude
Code loads it automatically.
Option C — Raw HTTP from any language
Every tool is a plain REST endpoint. Minimal stdlib Python client (no extra deps) in
tools/freecad_client.py:
from tools.freecad_client import FreeCADClient
fc = FreeCADClient() # defaults to http://127.0.0.1:8000
fc.tool("POST", "/document/create", name="demo")
fc.tool("POST", "/shapes/add_box", name="b", length=50, width=30, height=10)
info = fc.tool("POST", "/model/get_shape_info", shape_name="b")
print(info["volume_mm3"]) # 15000.0
The core modeling loop
create_document
→ add primitives (add_box / add_cylinder / add_cone / add_wing)
→ boolean ops (boolean_union / boolean_cut / make_hole)
→ inspect (get_shape_info / bounding_box / check_interference)
→ finish (fillet_edges / chamfer_edges)
→ export (export_step)
→ validate (validate_step) ← aim for is_clean == true, solid_count == 1
See AGENTS.md for the full endpoint reference, edge-selector rules, the in-place
boolean-update pattern, and the mechanical integration checks (interference / clearance) that
catch spatial errors before export.
Finite-element analysis (optional)
With Elmer and Gmsh installed, the /fem/* endpoints and the elmer_*.py runners cover a broad
tutorial set — elasticity, heat, electrostatics, magnetostatics, fluid flow, eigenmodes, and
more. These are heavier and require the external Elmer/Gmsh toolchain; the CAD tools work without
them.
Repository layout
| Path | What it is |
|---|---|
main.py |
FastAPI tool server — all HTTP endpoints |
session.py |
FreeCAD session logic (geometry, booleans, export, FEM) |
models.py |
Typed Pydantic request/response models |
mcp_server.py |
MCP bridge — exposes the tools to MCP clients |
freecad_bridge.py |
Headless FreeCAD discovery/import helper |
AGENTS.md / CLAUDE.md |
Agent operating manual (identical content) |
elmer_*.py |
Standalone Elmer FEM tutorial runners |
tools/ |
Client helper, test suite, integration-check hook, example scripts |
start.bat / start.sh |
Launchers (use FreeCAD's bundled Python) |
License
MIT.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。