cad-cae-copilot

CAD/CAE Copilot is an AI-native CAD, CAE and CAX workbench. An MCP-capable agent writes real build123d and OpenCASCADE geometry, exports STEP, STL and GLB, names the parts, exposes stable topology pointers, runs a deterministic critique, and can continue into CAE, with all artifacts preserved in one reproducible .aieng package. The user brings their own MCP client, such as Claude Code, Codex, Copilot or Cursor, which carries its own model access; the backend itself needs no API key.

The workflow runs from a natural-language engineering spec through the agent, the MCP tool layer, CAD modeling and editing, CAE setup and simulation, result evaluation and closed-loop optimization. A human-in-the-loop trust and approval layer gates every mutation, a credibility tier is stamped on every result, and the CAE path keeps material, loads, constraints, mesh and solver artifacts beside the CAD model. The project also ships a machine-readable analytical FEA benchmark corpus that compares solver metrics against documented closed-form references.

The README positions the tool against picture-only text-to-CAD: the output is editable, verifiable CAD with named parts and editable parameters, not a rendered image. Docker is the recommended first install because it pins the build123d, OpenCASCADE and CalculiX stack.