The text-to-cad repository is a JavaScript-based library that provides agent skills for computer-aided design, computer-aided engineering, and computer-aided manufacturing workflows. Hosted at texttocad.dev, the project enables AI agents to generate, inspect, source, and process CAD and robotics artifacts through a collection of focused, specialized skills rather than monolithic tools.

The library includes eleven distinct skills covering different aspects of the design and fabrication pipeline. The CAD skill creates and edits CAD models from plain-language or image requests, outputting STEP files with options to export to STL, 3MF, and GLB formats. The CAD Viewer skill provides local browser previews for CAD files, G-code, and robot description files. The step.parts skill sources off-the-shelf STEP components like screws, bearings, motors, and connectors. For 2D work, the DXF skill generates drawings including profiles, templates, gaskets, and cut layouts from Python sources or CAD geometry. Robot-focused skills include URDF for writing robot structure files with links, joints, limits, inertials, and meshes; SRDF for adding MoveIt planning groups, end effectors, poses, and collision rules to URDF files; and SDF for creating simulator models and worlds with frames, physics, sensors, and lights. The SendCutSend skill validates DXF and STEP files before upload to the SendCutSend fabrication service. The G-code skill slices supported mesh files into validated, printer-profiled FDM G-code using real slicer command-line interfaces. The Bambu Labs skill performs dry-runs, uploads, and cautiously starts local Bambu Lab print jobs from validated G-code. An experimental Implicit CAD skill creates browser-native implicit CAD models using GLSL signed-distance fields with CAD Viewer raymarch rendering.

Installation occurs through the Skills CLI as the preferred path, which installs individual skills directly for supported agents. Provider-native plugin installations are also available for Codex and Claude Code, with Codex supporting the repository-root plugin from version 0.142.0 onward. The repository distinguishes between production use from the main branch, which contains generated skill outputs needed by provider installers, and development work from the develop branch where contributors should open pull requests.

The project includes ten benchmark scenarios demonstrating the CAD skill's capabilities across mechanical engineering tasks ranging from simple rectangular calibration blocks to complex assemblies like planetary gear stages and spiral staircases. These benchmarks show the system generating geometry from natural language prompts, with each benchmark stored as markdown documentation accompanied by animated GIF previews. The repository uses Git LFS for storing heavyweight assets in the assets and benchmarks directories while excluding those trees from default LFS pulls to keep lightweight clones manageable. Benchmark markdown files remain in normal Git for readable diffs. The project maintains contribution guidelines in CONTRIBUTING.md that specify the develop branch as the target for pull requests and document the symlink workflow for local development.