JustusBraitinger/Autodesk-Fusion-360-MCP-Server

Fusion MCP Integration

https://github.com/user-attachments/assets/46c8140e-377d-4618-a304-03861cb3d7d9

🎯 About

Fusion MCP Integration bridges AI assistants with Autodesk Fusion 360 through the Model Context Protocol (MCP). This enables:

• ✨ Conversational CAD - Create 3D models using natural language
• 🤖 AI-Driven Automation - Automate repetitive modeling tasks
• 🔧 Parametric Control - Dynamically modify design parameters
• 🎓 Accessible CAD - Lower the barrier for non-CAD users

Note: This is designed as an assistive tool and educational project, not a replacement for professional CAD workflows. Projects like this can assist people with no experience in CAD workflows.

Goal: Enable conversational CAD and AI-driven automation in Fusion.

Setup

I highly recommend to do everything inside Visual Studio Code or an other IDE

Requirements

Clone Repository

git clone https://github.com/JustusBraitinger/FusionMCP

Important: Do NOT start the Add-In yet.

Install Python Dependencies

cd Server
python -m venv venv

Activate venv

Windows PowerShell

.\venv\Scripts\Activate

Install packages

pip install -r requirements.txt
pip install "mcp[cli]"

Installing the MCP Add-In for Fusion 360

cd ..
python Install_Addin.py

Connect to Claude

The most simple way to add the MCP-Server to Claude Desktop is to run following command:

cd Server
uv run mcp install MCP_Server.py

The output should be like this:

[11/13/25 08:42:37] INFO     Added server 'Fusion' to Claude config
                    INFO     Successfully installed Fusion in Claude app                                                                                                                                                               

Alternative

Modify Claude Config

In Claude Desktop go to:
Settings → Developer → Edit Config

Add this block (change the path for your system):

{
  "mcpServers": {
    "FusionMCP": {
      "command": "uv",
      "args": [
        "--directory",
        "C:\\Path\\to\\FusionMCP\\Server",
        "run",
        "MCP_Server.py"
      ]
    }
  }
}

Note: Windows paths require double backslashes \\

Using the MCP in Claude

1. Restart Claude if needed (force close if not visible)
2. Click ➕ Add (bottom left of chat)
3. Select Add from Fusion
4. Choose a Fusion MCP prompt

Use MCP in VS Code (Copilot)

Create or edit the file:

%APPDATA%\Code\User\globalStorage\github.copilot-chat\mcp.json

Paste:

{
  "servers": {
    "FusionMCP": {
      "url": "http://127.0.0.1:8000/sse",
      "type": "http"
    }
  },
  "inputs": []
}

Alternative Setup in VS Code

1. Press CTRL + SHIFT + P → search MCP → choose:
2. Add MCP
3. HTTP
4. Enter:
5. Name your MCP FusionMCP!!

http://127.0.0.1:8000/sse

Try It Out 😄

Activate the Fusion Addin inside Fusion

Configured in VS-Code:

Start the server:

python MCP_Server.py

Then type

/mcp.FusionMCP

Now you will see a list of predetermined Prompts.

Configured in Claude

Just open Claude, an ask for the FusionMCP

🛠️ Available Tools

✏️ Sketching & Creation Tools

⚙️ Feature & Modification Tools

📏 Analysis & Control

💾 Export

Architecture

Server.py

• Defines MCP server, tools, and prompts
• Handles HTTP calls to Fusion add-in

MCP.py

• Fusion Add-in
• Because the Fusion API is not thread-safe, this uses:
  • Custom event handler
  • Task queue

Why This Architecture?

The Fusion 360 API is not thread-safe and requires all operations to run on the main UI thread. Our solution:

1. Event-Driven Design - Use Fusion's CustomEvent system
2. Task Queue - Queue operations for sequential execution
3. Async Bridge - HTTP server handles async MCP requests

Security Considerations 🔒

• Local execution → safe by default
• Currently HTTP (OK locally, insecure on networks)
• Validate tool inputs to avoid prompt injection
• Real security depends on tool implementation

This is NOT

• ❌ A production-ready tool
• ❌ A replacement for professional CAD software
• ❌ Suitable for critical engineering work
• ❌ Officially supported by Autodesk

This IS

• ✅ A proof-of-concept
• ✅ An educational project
• ✅ A demonstration of MCP capabilities
• ✅ A tool for rapid prototyping and learning

This is a proof-of-concept, not production software.

If you want it to build yourself

• Use Websocket instead of plain HTTP
• Find a way to tell the llm the retarded faceindecies and body names

Contact

justus@braitinger.org