AI-Powered CAD System (MCP + Eyeshot)

This repository contains the hackathon prototype for an AI-driven CAD application utilizing the Model Context Protocol (MCP) and devDept Eyeshot SDK.

Architecture layers:

1. frontend: React + Three.js interface combining a Chat Assistant with a 3D Canvas.

2. backend-mcp: Python FastAPI server handling the MCP routing logic and tool execution.

3. llm-service: Python FastAPI server responsible for talking to the LLM (Gemini/GPT) to map natural language to CAD operations.

4. cad-engine: C# ASP.NET Core server utilizing the devDept Eyeshot SDK to perform CAD operations headlessly.

5. shared: A set of common schemas mapping commands across the layer divides.

Related MCP server: Fusion360 LLM Assistant

Setup

1. Copy .env.example to .env.

2. Assign your secret keys:

   • GEMINI_API_KEY: Your Gemini API key from Google AI Studio.

   • EYESHOT_LICENSE_KEY: Your Eyeshot production license key.

3. IMPORTANT: NEVER commit your .env file to the repository. It is already included in .gitignore to prevent accidental credential leaks.

Requirements

• Node.js (for the frontend)

• Python 3.10+ (for MCP and LLM services)

• .NET 8 SDK (for the CAD engine)

Command Flow example

• Prompt: "Load the sample.step model"

• MCP router sends to LLM service.

• LLM Service returns: {"action": "load_model", "file_path": "sample.step"}

• MCP router dispatches HTTP POST to CAD Engine with payload.

• CAD Engine executes the command and yields the modified state.