Fusion 360 MCP
Enables AI-powered parametric CAD design in Autodesk Fusion 360 through natural language commands. Supports multiple AI backends (Ollama, OpenAI, Gemini, Claude) with intelligent routing and safety validation for geometric operations.
README
Fusion 360 MCP - Multi-Model AI Integration
FusionMCP is a comprehensive Model Context Protocol (MCP) integration layer that connects Autodesk Fusion 360 with multiple AI backends (Ollama, OpenAI, Google Gemini, and Anthropic Claude) to enable AI-powered parametric CAD design through natural language.
🎯 Features
- 🤖 Multi-Model Support: Seamlessly switch between Ollama, OpenAI GPT-4o, Google Gemini, and Claude 3.5
- 🔄 Intelligent Routing: Automatic fallback chain when primary model fails
- 📐 Parametric Design: AI understands and generates parametric CAD operations
- 🛡️ Safety First: Built-in validation for dimensions, units, and geometric feasibility
- 💾 Context Caching: Conversation and design state persistence (JSON/SQLite)
- 🎨 Fusion 360 Integration: Native add-in for seamless workflow
- ⚡ Async Architecture: Fast, non-blocking operations with retry logic
- 📊 Structured Logging: Detailed logs with Loguru
📋 Table of Contents
- Architecture
- Installation
- Quick Start
- Configuration
- Usage Examples
- API Reference
- Model Comparison
- Development
- Troubleshooting
- Contributing
🏗️ Architecture
┌─────────────────────────────────────────────────────────────┐
│ Fusion 360 User │
│ ↓ │
│ ┌─────────────────────────┐ │
│ │ Fusion 360 Add-in │ │
│ │ - UI Dialog │ │
│ │ - Action Executor │ │
│ │ - Network Client │ │
│ └──────────┬──────────────┘ │
│ ↓ HTTP/REST │
│ ┌─────────────────────────┐ │
│ │ MCP Server (FastAPI) │ │
│ │ - Router │ │
│ │ - Schema Validation │ │
│ │ - Context Cache │ │
│ └──────────┬──────────────┘ │
│ ↓ │
│ ┌───────────────┴───────────────────┐ │
│ ↓ ↓ ↓ ↓ │
│ ┌─────────┐ ┌──────────┐ ┌────────┐ ┌──────────┐ │
│ │ Ollama │ │ OpenAI │ │ Gemini │ │ Claude │ │
│ │ (Local) │ │ API │ │ API │ │ API │ │
│ └─────────┘ └──────────┘ └────────┘ └──────────┘ │
│ │
│ System Prompt (FusionMCP Personality) │
│ ↓ │
│ Structured JSON Actions → Fusion 360 │
└─────────────────────────────────────────────────────────────┘
Component Overview
-
Fusion 360 Add-in (
fusion_addin/)- Python-based Fusion 360 add-in
- Captures user intent and design context
- Executes structured CAD actions
- Real-time UI feedback
-
MCP Server (
mcp_server/)- FastAPI-based REST server
- Routes requests to appropriate LLM
- Validates and normalizes responses
- Caches conversation history
-
LLM Clients (
mcp_server/llm_clients/)- Unified interface for all models
- Provider-specific implementations
- Automatic retry and error handling
-
System Prompt (
prompts/system_prompt.md)- Defines FusionMCP personality
- Enforces JSON output format
- Provides action schema templates
🚀 Installation
Prerequisites
- Python 3.11+ (for MCP server)
- Autodesk Fusion 360 (2025 version recommended)
- At least one LLM provider:
- Ollama (local, free)
- OpenAI API Key
- Google AI API Key
- Anthropic API Key
Step 1: Clone Repository
git clone https://github.com/yourusername/fusion360-mcp.git
cd fusion360-mcp
Step 2: Install Python Dependencies
# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Or install in development mode
pip install -e .
Step 3: Configure Environment
Create config.json from example:
cp examples/example_config.json config.json
Edit config.json with your API keys:
{
"ollama_url": "http://localhost:11434",
"openai_api_key": "sk-proj-...",
"gemini_api_key": "AIza...",
"claude_api_key": "sk-ant-...",
"default_model": "openai:gpt-4o-mini",
"mcp_host": "127.0.0.1",
"mcp_port": 9000
}
Alternative: Use environment variables (.env file):
OPENAI_API_KEY=sk-proj-...
GEMINI_API_KEY=AIza...
CLAUDE_API_KEY=sk-ant-...
Step 4: Install Fusion 360 Add-in
-
Copy
fusion_addin/folder to Fusion 360 add-ins directory:- Windows:
%APPDATA%\Autodesk\Autodesk Fusion 360\API\AddIns\ - macOS:
~/Library/Application Support/Autodesk/Autodesk Fusion 360/API/AddIns/
- Windows:
-
Rename to
FusionMCP:cp -r fusion_addin "/Users/YOUR_USER/Library/Application Support/Autodesk/Autodesk Fusion 360/API/AddIns/FusionMCP" -
Restart Fusion 360
-
Open Fusion 360 → Scripts and Add-Ins → Add-Ins tab → Select FusionMCP → Run
🎬 Quick Start
1. Start MCP Server
# Activate virtual environment
source venv/bin/activate
# Start server
python -m mcp_server.server
Expected output:
INFO | Logger initialized with level INFO
INFO | Cache initialized: json
INFO | System prompt loaded
INFO | Initialized MCP Router with providers: ['ollama', 'openai', 'gemini', 'claude']
INFO | MCP Server started on 127.0.0.1:9000
2. Test Server (Optional)
curl -X POST http://127.0.0.1:9000/mcp/command \
-H "Content-Type: application/json" \
-d '{
"command": "ask_model",
"params": {
"provider": "openai",
"model": "gpt-4o-mini",
"prompt": "Create a 20mm cube"
},
"context": {
"active_component": "RootComponent",
"units": "mm",
"design_state": "empty"
}
}'
3. Use in Fusion 360
- Open Fusion 360
- Click Scripts and Add-Ins → Add-Ins → FusionMCP → Run
- Click MCP Assistant button in toolbar
- Enter natural language command:
- "Create a 20mm cube"
- "Design a mounting bracket with 4 holes"
- "Make a cylindrical shaft 10mm diameter, 50mm long"
⚙️ Configuration
Full Configuration Options
{
// API Configuration
"ollama_url": "http://localhost:11434",
"openai_api_key": "sk-proj-...",
"gemini_api_key": "AIza...",
"claude_api_key": "sk-ant-...",
// Model Selection
"default_model": "openai:gpt-4o-mini",
"fallback_chain": [
"openai:gpt-4o-mini",
"gemini:gemini-1.5-flash-latest",
"ollama:llama3"
],
// Server Settings
"mcp_host": "127.0.0.1",
"mcp_port": 9000,
"allow_remote": false,
// Logging
"log_level": "INFO",
"log_dir": "logs",
// Caching
"cache_enabled": true,
"cache_type": "json", // or "sqlite"
"cache_path": "context_cache.json",
// Timeouts and Retries
"timeout_seconds": 30,
"max_retries": 3,
"retry_delay": 1.0,
// Available Models
"models": {
"ollama": {
"available": ["llama3", "mistral", "codellama"],
"default": "llama3"
},
"openai": {
"available": ["gpt-4o", "gpt-4o-mini", "gpt-4-turbo"],
"default": "gpt-4o-mini"
},
"gemini": {
"available": ["gemini-1.5-pro-latest", "gemini-1.5-flash-latest"],
"default": "gemini-1.5-flash-latest"
},
"claude": {
"available": ["claude-3-5-sonnet-20241022"],
"default": "claude-3-5-sonnet-20241022"
}
}
}
💡 Usage Examples
Example 1: Simple Geometry
Prompt: "Create a 20mm cube"
Generated Action:
{
"action": "create_box",
"params": {
"width": 20,
"height": 20,
"depth": 20,
"unit": "mm"
},
"explanation": "Creating a 20mm cubic box",
"safety_checks": ["dimensions_positive", "units_valid"]
}
Example 2: Complex Design
Prompt: "Design a mounting bracket 100x50mm with 4 M5 mounting holes"
Generated Action Sequence:
{
"actions": [
{
"action": "create_box",
"params": {"width": 100, "height": 50, "depth": 5, "unit": "mm"},
"explanation": "Create base plate"
},
{
"action": "create_hole",
"params": {"diameter": 5.5, "position": {"x": 10, "y": 10}, "unit": "mm"},
"explanation": "M5 clearance hole (10mm edge offset)"
},
// ... 3 more holes
],
"total_steps": 5
}
Example 3: Parametric Design
Prompt: "Create a shaft with diameter 2x of length"
{
"clarifying_questions": [
{
"question": "What is the shaft length?",
"context": "Need length to calculate diameter (diameter = 2 × length)",
"suggestions": ["50mm", "100mm", "Custom"]
}
]
}
📡 API Reference
Endpoints
POST /mcp/command
Execute MCP command.
Request Body:
{
"command": "ask_model",
"params": {
"provider": "openai",
"model": "gpt-4o-mini",
"prompt": "User prompt here",
"temperature": 0.7,
"max_tokens": 2000
},
"context": {
"active_component": "RootComponent",
"units": "mm",
"design_state": "empty"
}
}
Response:
{
"status": "success",
"message": "Action generated successfully",
"actions_to_execute": [...],
"llm_response": {...}
}
GET /health
Health check.
Response:
{
"status": "healthy",
"providers": ["ollama", "openai", "gemini", "claude"],
"cache_enabled": true
}
GET /models
List available models.
Response:
{
"models": {
"ollama": ["llama3", "mistral"],
"openai": ["gpt-4o", "gpt-4o-mini"],
"gemini": ["gemini-1.5-pro-latest"],
"claude": ["claude-3-5-sonnet-20241022"]
}
}
GET /history?limit=10
Get conversation history.
Response:
{
"conversations": [...],
"actions": [...]
}
Supported Actions
| Action | Description | Required Params |
|---|---|---|
create_box |
Create rectangular box | width, height, depth, unit |
create_cylinder |
Create cylinder | radius, height, unit |
create_sphere |
Create sphere | radius, unit |
create_hole |
Create hole | diameter, position, unit |
extrude |
Extrude profile | profile, distance, unit |
fillet |
Round edges | edges, radius, unit |
apply_material |
Apply material | material_name |
🔬 Model Comparison
| Feature | Ollama (Local) | OpenAI GPT-4o | Google Gemini | Claude 3.5 |
|---|---|---|---|---|
| Cost | Free | $$ | $ | $$$ |
| Speed | Fast | Medium | Fast | Medium |
| Offline | ✅ Yes | ❌ No | ❌ No | ❌ No |
| JSON Mode | Limited | ✅ Native | Good | Good |
| Reasoning | Good | Excellent | Very Good | Excellent |
| Geometry | Good | Very Good | Excellent | Very Good |
| Creative | Good | Excellent | Very Good | Good |
| Best For | Privacy, Offline | Creative designs | Spatial reasoning | Safety validation |
Recommended Workflows
- Creative Design: OpenAI GPT-4o → Claude (validation)
- Geometric Precision: Gemini → OpenAI
- Privacy-First: Ollama (all tasks)
- Cost-Optimized: Gemini Flash → Ollama (fallback)
🛠️ Development
Project Structure
fusion360-mcp/
├── mcp_server/ # MCP Server
│ ├── server.py # FastAPI app
│ ├── router.py # Request routing
│ ├── schema/ # Pydantic models
│ ├── llm_clients/ # LLM implementations
│ └── utils/ # Utilities
├── fusion_addin/ # Fusion 360 Add-in
│ ├── main.py # Entry point
│ ├── ui_dialog.py # UI components
│ ├── fusion_actions.py # Action executor
│ └── utils/network.py # Network client
├── prompts/ # System prompts
├── examples/ # Example configs
├── tests/ # Test suite
├── requirements.txt # Dependencies
└── README.md # This file
Running Tests
# Run all tests
pytest tests/ -v
# Run specific test file
pytest tests/test_mcp_server.py -v
# Run with coverage
pytest tests/ --cov=mcp_server --cov-report=html
Adding New LLM Provider
- Create client in
mcp_server/llm_clients/new_provider_client.py:
class NewProviderClient:
async def generate(self, model, prompt, system_prompt, temperature, max_tokens):
# Implementation
return {
"provider": "new_provider",
"model": model,
"output": "...",
"json": {...},
"tokens_used": 123
}
- Register in
router.py:
if config.new_provider_api_key:
self.clients["new_provider"] = NewProviderClient(...)
Code Style
- PEP8 compliant
- Type annotations required
- Docstrings for all functions/classes
- Async/await for I/O operations
🐛 Troubleshooting
Common Issues
1. Server Won't Start
Error: Address already in use
Solution: Change port in config.json:
{"mcp_port": 9001}
2. Fusion Add-in Not Visible
Solution:
- Verify add-in is in correct folder
- Check
FusionMCP.manifestexists - Restart Fusion 360
- Check Scripts and Add-Ins → Add-Ins tab
3. API Key Errors
Error: 401 Unauthorized
Solution:
- Verify API key in
config.json - Check key has proper permissions
- Try environment variables instead
4. Ollama Connection Failed
Error: Connection refused
Solution:
# Check Ollama is running
ollama list
# Start Ollama service
ollama serve
5. JSON Parsing Errors
Solution:
- Check system prompt is loaded
- Verify model supports JSON mode
- Use temperature < 0.8 for better structure
- Enable
json_mode=Truein OpenAI client
Debug Mode
Enable verbose logging:
{"log_level": "DEBUG"}
Check logs in logs/mcp_server.log
Health Check
# Check server health
curl http://127.0.0.1:9000/health
# List available models
curl http://127.0.0.1:9000/models
# View conversation history
curl http://127.0.0.1:9000/history?limit=5
🧪 Testing the System
Manual CLI Test
curl -X POST http://127.0.0.1:9000/mcp/command \
-H "Content-Type: application/json" \
-d @examples/example_command.json
Python Test Script
import requests
command = {
"command": "ask_model",
"params": {
"provider": "openai",
"model": "gpt-4o-mini",
"prompt": "Create a 10mm cube"
},
"context": {
"units": "mm",
"design_state": "empty"
}
}
response = requests.post("http://127.0.0.1:9000/mcp/command", json=command)
print(response.json())
🤝 Contributing
Contributions welcome! Please:
- Fork the repository
- Create feature branch (
git checkout -b feature/amazing-feature) - Commit changes (
git commit -m 'Add amazing feature') - Push to branch (
git push origin feature/amazing-feature) - Open Pull Request
Development Setup
# Install dev dependencies
pip install -e ".[dev]"
# Install pre-commit hooks
pre-commit install
# Run linting
ruff check mcp_server/
black mcp_server/
📄 License
MIT License - see LICENSE file
🙏 Acknowledgments
- Autodesk Fusion 360 API
- FastAPI framework
- Anthropic, OpenAI, Google for LLM APIs
- Ollama for local LLM support
📞 Support
- Issues: GitHub Issues
- Discussions: GitHub Discussions
- Documentation: Wiki
🗺️ Roadmap
- [ ] WebSocket streaming for real-time chat
- [ ] Vision model support (CAD screenshot analysis)
- [ ] Multi-agent orchestration
- [ ] Generative Design API integration
- [ ] Geometry export to Markdown/docs
- [ ] Fusion 360 UI palette integration
- [ ] 3D preview before execution
- [ ] Undo/redo action history
- [ ] Cloud deployment support
Built with ❤️ for the Fusion 360 and AI community
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。