1 / 3
【Overview】 agentcad is an open-source MCP server and CLI that gives a coding agent a real CAD feedback loop. The agent writes a build123d Python script, agentcad executes it, and the result comes back as a versioned STEP file together with geometric metrics such as dimensions, volume, face count and validity. On every successful run it also returns PNG renders, including an automatic four-view composite, so the geometry can be checked visually rather than guessed at. The problem it addresses is that a coding agent can write CAD code but cannot see or measure what that code produced. agentcad closes the loop from both sides: static checks catch syntax errors, a missing show_object() call and unresolvable imports before the CAD engine ever starts, and after a run the agent can measure feature sizes off the finished file, check the geometry against a written spec, inspect a topology report, and diff the new version against the previous success with side-by-side renders, overlay images and a solid-material volume diff. Meshes in STL, GLB and OBJ can be exported for 3D printing and web viewers, with GLB coloring each solid separately. It runs locally, is free and requires no signup. It suits developers already working with coding agents such as Claude Code, Cursor, Codex or any agent that can run shell commands in a folder, people with no CAD experience who want to describe a part in plain English, makers who need printable STL or STEP output straight out of an AI session, and teams iterating on multi-part assemblies that have to fit together. 【Key features】 • Execute: Runs build123d Python scripts and returns a versioned STEP file plus geometric metrics such as dimensions, volume, face count and validity. • Render: Produces PNG views of the geometry from any angle, with an automatic four-view composite preview on every successful run. • Export: Writes STL, GLB and OBJ meshes for 3D printing and web viewers, with GLB coloring each solid separately. • Validate: Static pre-execution checks catch syntax errors, a missing show_object(), and unresolvable imports before the CAD engine starts. • Measure and spec check: Reads feature sizes back off the finished CAD file and checks geometry against a written spec so the agent can catch mistakes and revise. • MCP integration: Turns every agentcad command into a native agent tool with typed parameters in and structured JSON out, keeping the CAD engine loaded between runs. 【Use cases】 1. Design a phone stand: an angled cradle that holds a phone at 60 degrees. 2. Model a coaster — 60mm square, 4mm thick, with rounded corners and a shallow dish — then export it as STL for printing. 3. Build a 234-part 3D reconstruction of the 1903 Wright Flyer from blueprints across 19 numbered versions. 【Pros and cons】 Pros: - The validation step runs before the CAD engine is invoked, so agents that tend to emit broken scripts get fast feedback instead of slow failures. - Renders plus the version-to-version diff give a reviewer something concrete to look at, which makes this a good fit for teams where nobody wants to open a CAD package. - Because output arrives as STEP and standard meshes, the results slot into an existing printing or web pipeline without extra conversion work. - MCP integration is the real win for agent users: the CAD engine stays loaded between runs, so iteration on a single part does not pay startup cost each time. Cons: - Execute requires Python 3.10 to 3.12; the OpenCascade bindings agentcad needs do not exist for 3.13 yet. - Diff always compares against the previous successful version only, so older baselines cannot be selected. - Renders default to 800x600 with MSAA at 0, and higher quality has to be requested explicitly. - Most agents only pick up a newly added MCP server after a restart. 【FAQ】 Q: Do I need CAD experience to use agentcad? A: No. You describe the part in plain English and the coding agent writes the build123d script; agentcad runs it and returns the STEP file, metrics and renders for verification. Q: Which Python version does agentcad need? A: Execute supports Python 3.10 to 3.12. The OpenCascade bindings it depends on are not yet available for 3.13. Q: What does a successful run give back? A: A versioned STEP file, geometric metrics such as dimensions, volume, face count and validity, PNG renders including a four-view composite, and mesh exports in STL, GLB or OBJ when you ask for them.
free
WPS AI is an AI assistant built into WPS Office that helps users write documents, generate presentation outlines and…
Baidu Comate is an AI coding assistant built by Baidu on its ERNIE model. It goes beyond simple autocomplete: it…
UniConverter is an AI media conversion and enhancement tool for video, audio, and images. It combines everyday…
Added
11/18/2026
Views
24
Clicks
5
6 similar tools to consider instead of agentcad