A new open-source project called text-to-cad is gaining rapid traction as one of the first comprehensive agent skill libraries designed to give AI coding assistants direct control over physical manufacturing and design workflows.
The library, created by developer Jake, provides a modular collection of skills that let agents like Claude Code, OpenAI Codex, and Grok Build generate CAD geometry, slice 3D prints for fabrication, write robot description files, and validate manufacturing outputs — all from natural-language requests.
With 14.8k GitHub stars accumulated within weeks of its public launch, text-to-cad is tapping into growing demand from engineers and hobbyists who want AI agents to handle the full cycle from idea to physical object.
The library ships eleven distinct skills organized around different stages of the CAD-to-manufacturing pipeline. The core CAD skill converts plain-language or image-based requests into parametric models, with STEP as the primary output format and exports to STL, 3MF, and GLB. A URDF skill writes robot structure files with links, joints, limits, inertials, and meshes, while the SRDF skill layers MoveIt planning groups, end effectors, and collision rules on top.
For fabrication, the DXF skill creates 2D cutting layouts, the G-code skill slices mesh files into validated FDM printer output using real slicer CLIs, and the DfAM Check skill measures printability metrics like wall thickness, overhangs, and support volume. A Bambu Labs skill handles dry-runs, uploads, and print initiation for Bambu Lab printers.
The project has also released provider-native plugin installs for Codex (version 0.142.0+), Claude Code, and Grok Build, plus a generic Skills CLI install path via npx. The project describes itself as still actively developed, with MIT licensing.
What makes text-to-cad notable is not just the breadth of manufacturing skills but the integration architecture. Each skill is designed as a self-contained workflow that an agent can invoke — a CAD Viewer skill provides local browser previews, while step.parts searches an off-the-shelf parts database for screws, bearings, motors, and connectors that agents can source directly.
The project represents a growing class of tools positioning AI agents not just as code generators but as bridges between digital intent and physical production. Similar efforts are emerging in CNC machining (CloudNC raised $20M this week for AI-driven CNC programming) and electronics design, suggesting a broader trend of AI agents moving from software-only domains into the physical world.
