McpCAD: AI-Powered CAD for VS Code

McpCAD is an AI-native 3D modeling system for VS Code. It provides an MCP server and a high-performance Open CASCADE (OCCT) viewport, enabling AI agents like Antigravity or Claude to see, measure, debug, and build 3D geometry directly in your workspace.

🚀 Why McpCAD?

Most AI agents are "blind" to 3D geometry. McpCAD bridges this gap by providing tools specifically designed for LLMs to:

• Visualize: Render complex B-Rep assemblies using a bundled Three.js viewport.
• Inspect: Query volumes, distances, bounding boxes, and face counts.
• Debug: Capture high-resolution multi-angle orthographic screenshots for visual feedback.
• Generate: Compile Python-based CadQuery scripts into concrete 3D entities.

✨ Key Features

• ⚡ Agent-Optimized: Tools are designed for LLMs, with flat schemas and descriptive metadata.
• 📐 B-Rep Edges: Explicitly renders arcs and wires, making it easy to debug geometry that lacks faces.
• 📸 Multi-Angle Capture: Generate 9-angle overview montages or specific zoomed views.
• 🌳 Assembly Tree: Interactive sidebar with visibility toggles and entity type identification.
• ✂️ Section Cuts: Dynamic X, Y, Z clipping planes for internal inspection.

🛠️ Getting Started

1. Install: Install the mcpcad extension from the VS Code Marketplace (or load the VSIX).
2. Setup Python: The extension will automatically help you set up a Python environment with cadquery and casadi.
3. Open Viewport: Run the command McpCAD Viewer.
4. Agent Integration: Ask your AI assistant (like Antigravity or Claude) to "Please design a ball bearing using McpCAD" or "Export a STEP of the model".

📦 Dependencies

McpCAD is built on the shoulders of giants:

• Open CASCADE: The world-class B-Rep kernel.
• CadQuery: A powerful nested-function Python library for parametric CAD.
• Three.js: For high-performance web-based 3D rendering.
• Model Context Protocol: For seamless AI instrument connectivity.

❓ F.A.Q.

Q: It's not installing, or it's taking forever to start.
A: The extension automatically prepares a Python environment and installs CadQuery on the first run. This process can take several minutes. Occasionally, a second restart is required for all components to sync. If the MCP server doesn't appear, please restart VS Code, re-open the McpCAD Viewport, and refresh your MCP tools.

Q: My AI agent says it can't "see" or render the model.
A: Ensure the McpCAD Viewport is open in VS Code (Run the McpCAD Viewer command). The agent interacts with the live viewport to capture screenshots and metadata.

Q: Why does my agent plan a complex assembly but then only generate a simple part?
A: AI agents often need a bit of "muscle memory" with new CAD APIs. We recommend letting your agent design a few simple standalone components first and saving those .py scripts in your workspace. Once the agent has these local examples to reference, it will be much more successful at "one-shotting" complex assemblies by building on its previous work.

⚖️ License

McpCAD is licensed under the Apache License 2.0. See the LICENSE file for details.

Created (prompted into existence) by Roope Kuisma © 2026 Taika3D Oy