What is text-to-cad?

Text-to-CAD, branded as CAD Skills, is a collection of skills that plug into AI coding agents like Claude Code, Codex, and Gemini CLI, giving them the ability to generate, export, and review CAD models, robot descriptions, and hardware design files. Rather than being a standalone CAD application, it extends the capabilities of coding agents you already use so they can produce real, source controlled design files from a written description. The core CAD skill lets an agent generate and export files in formats including STEP, STL, 3MF, DXF, and GLB, along with URDF, SDF, and SRDF robot description files used in robotics simulation and motion planning tools like MoveIt2. A companion CAD Explorer lets you visually inspect the generated geometry and robot descriptions, and a stable reference syntax lets you point the agent back at specific parts of a model for precise follow up edits. Additional skills cover finding and downloading off the shelf STEP parts, such as screws and bearings, from a hosted catalog, and preparing upload ready files for the SendCutSend fabrication service. Installing the skills is done through per agent scripts, for example a Claude Code install script that places the skills in the standard Claude Code skills directory, along with similar scripts for Codex, Gemini CLI, and OpenClaw. A universal installer command is also provided that can add the skills to any supported agent automatically. Generating actual CAD geometry also requires installing separate Python based CAD runtime dependencies, described in the setup instructions. The typical workflow is to describe a part or mechanism to your coding agent, let it edit the underlying source files, regenerate the specific export targets you need, then inspect the result in CAD Explorer before committing both the source and generated files together. The project includes a benchmark suite of test prompts with expected geometry outputs, and heavier assets are stored using Git LFS to keep normal clones lightweight. It is released under the MIT license.