OCP Viewer MCP Server

An MCP (Model Context Protocol) server that captures screenshots from the OCP CAD Viewer, allowing AI assistants like Cursor to "see" your 3D CAD models.

What it does

When working with CAD tools like CadQuery or Build123d, you can display models in the OCP CAD Viewer using show(). This MCP server lets you ask an AI assistant to capture what's currently displayed, enabling visual feedback and AI-assisted CAD development.

Example workflow:

1. You're designing a part in a Jupyter notebook with CadQuery
2. You run show(my_part) to display it in the OCP viewer
3. You ask the AI: "capture the current view - does this geometry look correct?"
4. The AI captures a screenshot and can analyze the 3D model visually

Installation

pip install ocp-viewer-mcp

Requirements:

• Python 3.10+
• OCP CAD Viewer VS Code/Cursor extension
• ocp-vscode Python package (for show() function)

Configuration

Cursor

Add to your ~/.cursor/mcp.json:

{
  "mcpServers": {
    "ocp-viewer": {
      "command": "python",
      "args": ["-m", "ocp_viewer_mcp.server"]
    }
  }
}

If you're using a virtual environment (recommended), specify the full Python path:

{
  "mcpServers": {
    "ocp-viewer": {
      "command": "/path/to/your/venv/bin/python",
      "args": ["-m", "ocp_viewer_mcp.server"]
    }
  }
}

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "ocp-viewer": {
      "command": "python",
      "args": ["-m", "ocp_viewer_mcp.server"]
    }
  }
}

Usage

1. Open your CadQuery/Build123d project in Cursor or VS Code
2. Start the OCP CAD Viewer (Cmd+Shift+P → "OCP CAD Viewer: Open Viewer")
3. Display a model using show(your_model) in a notebook or script
4. Ask the AI to capture the view:
   • "Capture the current OCP view"
   • "Show me what the model looks like"
   • "Take a screenshot of the CAD viewer"

Tool Reference

capture_ocp_screenshot

Captures a screenshot of the OCP CAD Viewer.

Parameter	Type	Default	Description
port	integer	3939	The port where OCP viewer is running
wait_ms	integer	1000	Milliseconds to wait for rendering

Returns: A PNG screenshot as a base64-encoded image.

Troubleshooting

"Could not connect to OCP viewer"

• Make sure the OCP CAD Viewer is open in VS Code/Cursor
• Check that the viewer backend is running (look for "Viewer backend started" in terminal)
• Verify the port (default is 3939)

"ocp_vscode not installed"

The MCP server needs ocp-vscode installed in the same Python environment:

pip install ocp-vscode

Screenshot not capturing

• Make sure a model is displayed (run show(something) first)
• Try increasing wait_ms if the model is complex

Development

# Clone the repo
git clone https://github.com/dmilad/ocp-viewer-mcp.git
cd ocp-viewer-mcp

# Install with Poetry
poetry install

# Run locally
poetry run python -m ocp_viewer_mcp.server

Contributing

Contributions are welcome! Please open an issue first to discuss what you'd like to change.

License

MIT

Related Projects

• OCP CAD Viewer - The VS Code extension this MCP server integrates with
• CadQuery - Python parametric CAD scripting
• Build123d - Python CAD scripting with a builder pattern
• MCP - Model Context Protocol specification