text-to-cad: CAD superpowers for your agent

TL;DR: text-to-cad is a Python project with 16,887 stars and 1,743 forks whose entire description is six words: “Give your agent CAD superpowers.” It ships under MIT, carries topics for agents, CAD, mechanical engineering and robotics, and points at a demo site at texttocad.dev.

What the repository claims

The GitHub description is unusually blunt. There is no feature list, no benchmark, no comparison table, just the sentence above. That makes the metadata around it the most useful thing to read, because there is nothing else in the repo header to lean on.

Field	Value
Full name	earthtojake/text-to-cad
Description	Give your agent CAD superpowers.
Stars	16,887
Forks	1,743
Primary language	Python
License	MIT
Topics	agents, ai-agents, cad, mechanical-engineering, robotics
Homepage	https://www.texttocad.dev
Created	2026-04-22
Last push	2026-10-05
Owner type	User

A fork count of roughly one per ten stars is high for a library. That is the ratio you see when people clone a repository to read it, change it or run it on their own machine, rather than pulling a package from an index. For a project whose value proposition is handing CAD ability to an agent, that pattern fits: you want the source locally so the agent can read the same code you are reading.

What the topics narrow down

GitHub exposes five topics, and they do more to place the project than the description does:

• agents and ai-agents place it inside the coding-agent ecosystem rather than the general CAD one.
• cad and mechanical-engineering anchor the domain.
• robotics is the most selective of the five, since robotics teams are the ones doing agent orchestration and solid modelling in the same week.

Read together, the topics describe a tool aimed at people who already run an agent and already work with geometry, which is a much smaller audience than “CAD software” and a much more specific one.

How to size it up yourself

There is a companion site at texttocad.dev, but the repository is a single Python project, so a clone is the fastest way to judge scope before you commit time to it:

git clone https://github.com/earthtojake/text-to-cad.git
cd text-to-cad
ls

The topic list is also worth querying directly if you are scanning a domain rather than evaluating one repo, since topics are the only structured metadata the project exposes beyond language and license:

curl -s https://api.github.com/repos/earthtojake/text-to-cad \
  | jq '{topics, language, license: .license.spdx_id, stars: .stargazers_count}'

The MIT license is the other thing worth noting early. For a project that ships agent-facing tooling, permissive licensing means you can read it, adapt the prompts or helpers into your own agent setup, and redistribute that without an open-source obligation on your side.

When it fits

This is a project to watch rather than one to bet a production pipeline on yet. The signals that make it interesting are cheap to check: the MIT license is uncontroversial, Python is the dominant language for agent tooling, and the last push landed the same morning this post was written, so the project is actively moving.

What you are missing from the public header is the part that matters for evaluation. The description does not say which CAD kernel it targets, what geometry formats it reads or writes, whether it renders previews, or what an agent is actually handed at runtime. A wiki is enabled on the repository, so that documentation exists, but it is not in the repository metadata itself. Check the wiki and the live site before you assume it covers STEP or STL.

The robotics topic is the one to weight most heavily if you are deciding whether this is relevant. Teams outside mechanical CAD will find the framing interesting and the use case narrow.

Source and Accuracy Notes

• Project page: https://www.texttocad.dev
• Source repository: github.com/earthtojake/text-to-cad (verified via GitHub API html_url)
• License: MIT (verified via GitHub API license.spdx_id)
• Stars at time of writing: 16,887
• Forks at time of writing: 1,743
• Facts in this post come from the GitHub API repository endpoint only. The repository README was not fetched, so no feature, command or integration is claimed here beyond what the description, topics and language field state.