Text-to-CAD

Overview

Text-to-CAD bridges the gap between AI coding agents and physical manufacturing. Instead of writing CAD scripts by hand or learning complex parametric modeling tools, developers can describe what they want in natural language and get production-ready 3D models, engineering drawings, and manufacturing files. The project ships as a local MCP server, so any agent that speaks MCP can call its skills directly — no custom integration required.

Key Capabilities

The skill library covers the full design-to-fabrication pipeline. You can generate 3D models in industry-standard formats (STEP for CAD interchange, GLB for web viewers, STL and 3MF for 3D printing), produce dimensioned engineering drawings as PDFs, run design-for-manufacturing checks against sheet metal, CNC, and injection molding constraints, generate robot description files (URDF/SRDF) for robotics simulation, and slice models into G-code for 3D printing. The @cad[…] reference system is particularly clever — it lets agents address specific geometric elements (faces, edges, features) across turns, enabling iterative refinement without losing context about what was already built.

Use Cases

Hardware engineers and roboticists use Text-to-CAD to rapidly prototype mechanical parts and assemblies without leaving their terminal. Robotics teams generate URDF files for simulation pipelines. Makers use it to go from idea to 3D-printable G-code in a single agent session. The fabrication service integration (SendCutSend) even enables ordering manufactured parts directly from the agent workflow.

Considerations

Text-to-CAD requires Python 3.11+ and installs via the uv package manager. The underlying OpenCASCADE kernel is powerful but can be resource-intensive for complex assemblies. Generated models should always be reviewed by an engineer before sending to production manufacturing — AI-generated geometry can have subtle issues that matter at fabrication tolerances.