A CAD design tool for coding agents (free + open source)
Agentcad is an open-source MCP server and CLI that gives coding agents the ability to design, render, and export real CAD output. It falls into the category of AI-assisted CAD tools and is specifically built for developers who work with agents such as Claude Code, Cursor, or Windsurf. The core value is enabling these agents to autonomously generate valid, printable geometry without constant human oversight. The tool runs locally, requires no signup, and is completely free under the Apache-2.0 license. By integrating through MCP, agentcad acts as a seamless bridge between natural language prompts and concrete 3D models. This allows developers to hand off CAD tasks to their agents, streamlining the design process from concept to prototype.
The concrete problem agentcad solves is that AI agents often produce flawed geometry on the first attempt, requiring tedious manual debugging. Without a tight feedback loop, agents generate invalid shapes that cannot be printed. Agentcad provides a rapid iterative cycle where the agent runs a script and immediately receives geometric metrics and renderings. This allows the agent to catch its own mistakes before they reach the developer. Pre-execution validation checks catch errors in under 100 milliseconds, ensuring only valid geometry proceeds. As a result, developers spend less time babysitting AI outputs and more time trusting their agents to handle complex modeling tasks.
The first major feature group is execution and validation. Agentcad lets agents execute Python scripts using build123d primitives with `show_object` pre-injected. On each successful run, it produces versioned STEP files along with detailed metrics like volume, face count, dimensions, and validity status. The validation step performs pre-execution checks that catch syntax errors and logical issues quickly. This instant feedback enables the agent to iterate rapidly and correct flaws autonomously. By automating validation, agentcad ensures only robust models proceed to rendering and export, reducing wasted time on bad geometry.
The second major feature group is rendering and export. After a successful run, agentcad generates PNG previews from four angles: top, isometric front-right, isometric back-right, and isometric back-left. These visualizations allow both the agent and the human to inspect the design from multiple perspectives. For export, agentcad supports STL, GLB, and OBJ formats, covering 3D printing and web-based viewers. This means a design created by an AI agent can be directly used in a slicer or embedded in a webpage without extra conversion. The combination of render and export closes the loop from abstract code to tangible visual and physical outputs.
Publisher
A
admin
Launch Date2026-06-10
Pricingfree
Domain Authority
Ahrefs DR7/100