What languages and tools power text-to-cad?

The core package, cadgen, is written in Python and built on build123d and OpenCASCADE (via the OCP binding). Requires Python 3.11+ and uv for running.

What agents does text-to-cad support?

Claude Code, Claude Desktop, Codex, Cursor, Grok Build, and Gemini have dedicated plugin install paths. Any other agent with skills support can use the fallback Skills CLI install.

What file formats can it generate?

Primary CAD output is STEP, with export options to STL, 3MF, and GLB for 3D models, plus DXF for 2D drawings and PDF for dimensioned engineering drawings.

Does it require an internet connection?

Yes, at least for the first run, since cadgen's runtime downloads then. After that, it runs fully offline aside from optional daily update checks and telemetry, both of which can be disabled.

Is there a known issue on Windows?

Yes. On Windows 11 with Smart App Control enabled (the default on fresh installs), the unsigned OCP native module fails to load. The fix is to disable Smart App Control or run the skills under WSL.

Text-to-CAD: Agent skills that let AI coding

text-to-cad is a collection of agent skills that give AI coding assistants (Claude Code, Codex, Cursor, Gemini, Grok, and others) the ability to design and manipulate CAD models. It's aimed at developers, mechanical engineers, and robotics builders who want their agent to generate 3D parts, write robot description files, or produce manufacturing-ready drawings without leaving their chat or editor.

Under the hood it runs on build123d and OpenCASCADE (via the OCP Python binding), packaged as the cadgen CLI/MCP server distributed on PyPI. The plugin installs as a set of skills plus a local viewer server, so agents can both generate and display models inline.

Key features

Text-to-model generation: turns plain-language or image requests into CAD geometry, exporting STEP as the primary format with STL, 3MF, and GLB options.
Off-the-shelf part lookup: the step.parts skill finds STEP files for screws, bearings, motors, and connectors.
Engineering drawings: generates dimensioned PDF drawings from a part, including views, hidden lines, hole callouts, and a title block.
DXF export: creates 2D drawings like profiles, templates, gaskets, and cut layouts from Python sources or CAD geometry.
Robot description files: writes URDF (links, joints, limits, inertials, meshes) and SRDF (MoveIt planning groups, end effectors, poses, collision rules).
Local viewer: shows models as viewer cards inline in supported agent UIs, or as a browser link in terminal-style agents.
Fabrication service connections: links to 3D printing, sheet metal, and CNC fabrication workflows alongside design-for-manufacturing checks.
Privacy-conscious telemetry: usage stats and crash reports are opt-out, anonymized, and never include file contents, paths, or prompts.

Ideal use cases

This fits teams building hardware products or robots who want an agent to iterate on part geometry, export fabrication-ready files, or scaffold robot description files (URDF/SRDF) from a conversation instead of hand-writing XML. It's a good match for anyone already using Claude Code, Codex, Cursor, Gemini, or Grok and wanting CAD capability added as a plugin rather than a separate app.

It also suits quick mechanical prototyping: asking an agent to generate a bracket, gasket profile, or enclosure, check it for manufacturability, and produce a drawing for a machine shop, all from text.

It is not a fit if you need a full interactive CAD GUI for freehand sketching, or if your workflow depends on proprietary CAD formats beyond STEP/STL/3MF/GLB/DXF. It also won't help if your agent app has no plugin or skills support and you're unwilling to run the Skills CLI fallback. Teams needing signed native binaries on locked-down Windows 11 machines (Smart App Control on) will hit friction, since the underlying OCP wheel ships unsigned native code.

Installation

CAD workflows run through uv, so install it first and confirm with uv --version.

The simplest path is to ask your agent directly:

Install text-to-cad from https://github.com/earthtojake/text-to-cad

For manual setup, install commands differ by agent. For Claude Code:

claude plugin marketplace add earthtojake/text-to-cad#latest
claude plugin install text-to-cad@earthtojake

For Codex:

codex plugin marketplace add earthtojake/text-to-cad --ref latest
codex plugin add text-to-cad@earthtojake

For Cursor (clone directly, since Cursor also reads Claude Code's plugin):

git clone --depth 1 --branch latest https://github.com/earthtojake/text-to-cad ~/.cursor/plugins/local/text-to-cad

For Grok Build:

grok plugin install earthtojake/text-to-cad@latest --trust
grok plugin enable text-to-cad

For Gemini:

gemini extensions install https://github.com/earthtojake/text-to-cad --ref latest --consent --auto-update

For an agent without plugin support, install the skills directly via the Skills CLI:

npx skills add earthtojake/text-to-cad#latest

After install, restart the agent app. The first start downloads cadgen's runtime, so a network connection is required. Update and reinstall commands mirror the install commands for each agent (marketplace update/uninstall, plugin update/remove, or re-running npx skills add).

Telemetry is on by default (anonymized usage stats and crash reports, no file contents or prompts) and can be disabled with uvx cadgen telemetry off.