MCP Server for CAD - LLM-driven 3D Geometry Manipulation

A local-first, open-source MVP of an MCP (Model Context Protocol) type server that uses a cloud LLM (via Groq's free API running Llama 3.3) to translate natural language user commands into structured geometric operations for a simulated CAD engine.

Prerequisites

1. Python 3.9+

2. A free Groq API key — get one at https://console.groq.com/keys

Setup

1. Install the project dependencies:

   cd mcp-cad-server
   pip3 install -r requirements.txt

2. Start the FastAPI backend server with your Groq key:

   export GROQ_API_KEY='your-groq-api-key-here'
   python3 -m uvicorn app.main:app --reload

   The server runs by default on http://localhost:8000.

Architecture

• Groq Cloud API: Free LLM inference using Llama 3.3 70B.

• FastAPI / MCP Server: Orchestrates receiving user prompts, formatting them for the LLM, securely parsing the JSON output, and delegating instructions.

• Mock CAD Engine: Simulated geometry state handling dimensions and volumetric output based on structured commands.

Testing via CLI

You can easily interact with the running CAD Server through test_cli.py:

python3 test_cli.py "Increase hole diameter by 5mm"

Expected JSON response:

{
  "status": "success",
  "message": "Successfully modified hole to 5.0mm.",
  "data": {
    "feature": "hole",
    "new_value": 5.0,
    "unit": "mm"
  }
}

More examples:

python3 test_cli.py "Create a sphere"
python3 test_cli.py "Extrude the top face by 50mm"
python3 test_cli.py "What is the volume?"

Supported Mock CAD Functions

• create_shape (shape_type, dimensions) — Generates box, sphere, cylinder, cone, or torus

• modify_dimension (feature, value, unit)

• extrude (face, distance, unit)

• get_volume (unit)