vcad ships an MCP (Model Context Protocol) server that lets AI assistants create, inspect, and export 3D models on your behalf. Once configured, you can describe a part in natural language and the AI will call vcad tools directly -- no manual modeling required.
How It Works
The MCP server runs as a local process that exposes vcad's geometry engine as a set of tools. An MCP client (Claude Desktop, Cursor, or any compatible editor) launches the server automatically when needed and communicates with it over stdio. The AI sees tool descriptions like create_cad_document and inspect_cad, decides when to call them based on your conversation, and receives structured results back.
The server includes the full WASM kernel, so geometry evaluation happens locally. No data leaves your machine.
Claude Desktop
Edit the Claude Desktop configuration file. On macOS it lives at ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows at %APPDATA%\Claude\claude_desktop_config.json.
Add the vcad server to the mcpServers object:
{
"mcpServers": {
"vcad": {
"command": "npx",
"args": ["@vcad/mcp"]
}
}
}
Restart Claude Desktop. You should see a hammer icon in the input area indicating that MCP tools are available. If the icon doesn't appear, check that Node.js (v18+) is installed and that npx is on your PATH.
Cursor
Cursor uses the same configuration format. Open Settings, navigate to the MCP section, and add a new server:
{
"mcpServers": {
"vcad": {
"command": "npx",
"args": ["@vcad/mcp"]
}
}
}
After saving, Cursor will launch the vcad server whenever an agent session starts. The tools appear in the agent's tool list alongside file editing and terminal tools.
Claude Code
If you use Claude Code (the CLI), add the server to your project's .mcp.json or your global MCP config:
{
"mcpServers": {
"vcad": {
"command": "npx",
"args": ["@vcad/mcp"]
}
}
}
Verifying the Setup
The quickest test is to ask the AI to create something simple. In Claude Desktop, type:
Create a 30mm cube in vcad
Claude should call the create_cad_document tool with a cube primitive and return a confirmation that includes the IR document. If it also calls inspect_cad, you'll see volume (27000 mm3), bounding box, and triangle count in the response.
If the AI says it doesn't have CAD tools, the server likely failed to start. Run npx @vcad/mcp directly in a terminal to check for errors. The most common issue is a missing or outdated Node.js installation.
Available Tools
Once connected, the AI has access to the full vcad tool suite. Here is what it can do at a glance:
| Tool | Purpose |
|---|---|
create_cad_document | Build geometry from primitives, sketches, booleans, and assemblies |
create_cad_loon | Build geometry using the compact Loon text format |
export_cad | Export to STL or GLB |
inspect_cad | Query volume, surface area, bounding box, center of mass |
import_step | Import STEP files into vcad format |
open_in_browser | Generate a shareable vcad.io URL |
create_robot_env | Create a physics simulation from an assembly |
gym_step / gym_reset / gym_observe / gym_close | Control physics simulations |
Each tool is documented in detail in the MCP Tools reference.
What's Next
Now that the server is connected, move on to Creating Models with AI to learn how create_cad_document works and see what kinds of geometry an AI agent can produce.