agentcad
<p> A CAD design tool for coding agents (free + open source) </p> <p> <a href="https://www.producthunt.com/products/agentcad?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1165633?app_id=339">Link</a> </p>
Last verified:
What is agentcad?
Agentcad is an open-source MCP server and CLI tool that enables AI coding agents to design, render, and export real CAD output. It lets agents like Claude Code, Codex, and Cursor create 3D models by writing build123d or CadQuery Python scripts, then automatically runs them to check their own work before showing results to humans.
The tool provides a tight feedback loop for agents: execute scripts to produce versioned STEP files with geometric metrics, render PNG views from any angle for visual verification, export to STL/GLB/OBJ formats for 3D printing and web viewers, validate geometry with pre-execution checks under 100ms, inspect topology reports for debugging, and diff versions to track design iterations. This helps agents catch broken code, confirm geometry is watertight and dimensionally correct, and fix mistakes autonomously.
Agentcad is designed for human-agent collaboration and is ideal for engineers, CAD designers, 3D printing enthusiasts, and anyone using coding AI agents who needs manufacturable parts. It requires zero boilerplate imports since build123d primitives and show_object are pre-injected, and it integrates with Claude Code, Cursor, and Windsurf via MCP configuration.
The tool runs locally on your machine, supports Python 3.10-3.12, and outputs interactive viewers plus STEP/STL/GLB files ready to inspect, edit, or print. A notable case study shows one Claude session building a 1903 Wright Flyer with 234 parts using only prompts.
Key features include automatic PNG rendering from any angle, geometric metrics (dimensions, volume, face count, validity), sub-100ms validation, version diffing, topology inspection, and export formats optimized for 3D printing and web viewing.
agentcad pricing
Pricing model: Freemium
Completely free and open source under Apache-2.0 license. No paid plans, no subscription fees, no tier limitations. The tool requires no signup, has no account requirements, and runs entirely locally on your machine. All features including execute, render, export, validate, inspect, and diff are available without any payment.
agentcad pros
- Open source with Apache-2.0 license
- Completely free with no payment required
- No signup or account creation needed
- Runs locally on your machine for privacy
- Supports Python 3.10 through 3.12
- Zero boilerplate imports needed for scripts
- Pre-injected build123d primitives and helpers
- Sub-100ms pre-execution validation checks
- Automatic PNG rendering from any angle
- Exports STEP, STL, GLB, and OBJ formats
- Geometric metrics including dimensions and volume
- Version diffing to track design iterations
- Topology reports for debugging geometry
- MCP integration with Claude Code and Cursor
- Works with Windsurf coding agent
- Installable agent skill on skills.sh and ClawHub
- Supports CLI-only mode without MCP server
- uv tool installation option available
- Interactive viewer included with output files
- Agents catch their own mistakes autonomously
agentcad cons
- Only supports Python 3.10-3.12, not newer versions
- Requires Python 3.12 virtual environment setup
- MCP server requires .mcp.json configuration
- CLI-only mode lacks MCP server integration
- Pre-injected helpers may limit custom imports
- No cloud-based rendering or storage option
- Limited to build123d and CadQuery libraries
- No built-in GUI for manual editing
- uvx caches environment on first run causing delay
- Release updates require optional signup
Frequently asked questions about agentcad
What is agentcad?
Agentcad is an open-source MCP server and CLI that lets your coding agent design, render, and export real CAD output. It enables AI agents like Claude Code, Codex, and Cursor to create 3D models by writing build123d or CadQuery Python scripts and automatically checking their work.
Is agentcad free to use?
Yes, agentcad is completely free and open source under the Apache-2.0 license. There are no paid plans, subscription fees, or tier limitations. No signup or account creation is required, and it runs entirely locally on your machine.
Which Python versions does agentcad support?
Agentcad currently supports Python 3.10, 3.11, and 3.12. The recommended setup uses Python 3.12 with a virtual environment created via python3.12 -m venv .venv.
What AI agents work with agentcad?
Agentcad works with Claude Code, Codex, Cursor, and Windsurf. You can add MCP integration by configuring your .mcp.json file, or use the CLI-only mode without MCP server integration.
What file formats can agentcad export?
Agentcad exports STEP files (versioned), STL, GLB, and OBJ formats. STEP files are produced with geometric metrics, while STL/GLB/OBJ are optimized for 3D printing and web viewers.
How does agentcad help agents catch mistakes?
Agentcad provides a tight feedback loop: run scripts, render PNG views, inspect metrics, and fix errors. Agents read both the metrics (dimensions, volume, face count, validity) and renders from any angle, allowing them to catch broken code and confirm geometry is watertight before showing results to humans.
Do I need to write imports for build123d scripts?
No, agentcad requires zero boilerplate imports. build123d primitives, show_object, and agentcad edit helpers are pre-injected automatically, so you can directly use syntax like box = Box(10, 20, 5) and show_object(box).
What validation does agentcad provide?
Agentcad provides pre-execution validation checks that catch errors in under 100ms. It also produces geometric metrics including dimensions (x, y, z), volume, face count, and an is_valid boolean to confirm the geometry is watertight and dimensionally correct.
Can I track design iterations with agentcad?
Yes, agentcad includes diff functionality to compare versions and track design iteration. Every run produces versioned STEP files, allowing you to see how designs change between iterations and monitor the agent's progress.