
CAD MCP Server
Give your AI assistant the ability to inspect, measure, and compare 3D CAD models. Drop in a STEP file, ask engineering questions, get measured answers. Runs entirely on your machine β no cloud, no CAD license, no setup.
Why Engineers Use It
-
Zero setup. One command. No CAD software, licenses, Docker, or cloud API.
-
Runs locally. Your STEP files never leave your machine.
-
Deterministic answers. Backed by the Open CASCADE kernel, not LLM guessing.
-
Read-only. Inspects geometry without modifying anything.
What It Can Do
Tool Example question
inspect_step "What are the overall dimensions and volume?"
query_faces "Find all cylindrical faces. Which ones are holes vs bosses?"
query_edges "What's the smallest fillet radius on this part?"
measure_step "Check wall thickness around every hole. Flag anything below 2mm."
diff_step "What changed between revision A and revision B?"
The AI assistant interprets the measurements. You get engineering answers, not raw numbers.
Example Prompts
-
"Review this part for injection molding: check draft angles with +Z pull, measure minimum wall thickness, identify undercuts."
-
"Prepare a first-pass CNC plan: likely setups, drilling directions, features that drive cost."
-
"Compare these two revisions and flag what needs rechecking before tooling."
-
"Which holes are blind vs through? What's the depth of the blind hole?"
-
"Find the thinnest wall section on this part. Is it above the 1.5mm minimum?"
See docs/EXAMPLE_PROMPTS.md for more.
License
MIT. The bundled OCCT kernel uses LGPL-2.1. See THIRD_PARTY_NOTICES.md.
npx -y cad-mcp-serverQuick Start
npx -y cad-mcp-server
Add to your MCP client:
{
"mcpServers": {
"cad": {
"command": "npx",
"args": ["-y", "cad-mcp-server"]
}
}
}
Point your AI at any STEP file and ask questions like:
"Review this part before manufacturing. Check wall thickness, draft angles, and hole sizes. Flag anything that violates standard DFM rules."
Requirements
-
Node.js 22+
-
STEP files (export from SolidWorks, FreeCAD, Fusion 360, CATIA, or any CAD system)
No common issues documented yet. If you hit a problem, the repository's GitHub Issues page is the best place to look.