YOLO MCP Service
A powerful YOLO (You Only Look Once) computer vision service that integrates with Claude AI through Model Context Protocol (MCP). This service enables Claude to perform object detection, segmentation, classification, and real-time camera analysis using state-of-the-art YOLO models.
Features
- Object detection, segmentation, classification, and pose estimation
- Real-time camera integration for live object detection
- Support for model training, validation, and export
- Comprehensive image analysis combining multiple models
- Support for both file paths and base64-encoded images
- Seamless integration with Claude AI
Setup Instructions
Prerequisites
- Python 3.10 or higher
- Git (optional, for cloning the repository)
Environment Setup
- Create a directory for the project and navigate to it: ```bash mkdir yolo-mcp-service cd yolo-mcp-service ```
- Download the project files or clone from repository: ```bash # If you have the files, copy them to this directory # If using git: git clone https://github.com/GongRzhe/YOLO-MCP-Server.git . ```
- Create a virtual environment: ```bash # On Windows python -m venv .venv # On macOS/Linux python3 -m venv .venv ```
- Activate the virtual environment: ```bash # On Windows .venv\Scripts\activate # On macOS/Linux source .venv/bin/activate ```
- Run the setup script: ```bash python setup.py ``` The setup script will: - Check your Python version - Create a virtual environment (if not already created) - Install required dependencies - Generate an MCP configuration file (mcp-config.json) - Output configuration information for different MCP clients including Claude
- Note the output from the setup script, which will look similar to: ``` MCP configuration has been written to: /path/to/mcp-config.json MCP configuration for Cursor: /path/to/.venv/bin/python /path/to/server.py MCP configuration for Windsurf/Claude Desktop: { "mcpServers": { "yolo-service": { "command": "/path/to/.venv/bin/python", "args": [ "/path/to/server.py" ], "env": { "PYTHONPATH": "/path/to" } } } } To use with Claude Desktop, merge this configuration into: /path/to/claude_desktop_config.json ```
Downloading YOLO Models
Before using the service, you need to download the YOLO models. The service looks for models in the following directories:
- The current directory where the service is running
- A `models` subdirectory
- Any other directory configured in the `CONFIG["model_dirs"]` variable in server.py
Create a models directory and download some common models:
For Windows PowerShell users:
Configuring Claude
To use this service with Claude:
- For Claude web: Set up the service on your local machine and use the configuration provided by the setup script in your MCP client.
- For Claude Desktop: - Run the setup script and note the configuration output - Locate your Claude Desktop configuration file (the path is provided in the setup script output) - Add or merge the configuration into your Claude Desktop configuration file - Restart Claude Desktop
Using YOLO Tools in Claude
1. First Check Available Models
Always check which models are available on your system first:
2. Detecting Objects in an Image
For analyzing an image file on your computer:
You can also specify a different model:
3. Running Comprehensive Image Analysis
For more detailed analysis that combines object detection, classification, and more:
4. Image Segmentation
For identifying object boundaries and creating segmentation masks:
5. Image Classification
For classifying the entire image content:
6. Using Your Computer's Camera
Start real-time object detection using your computer's camera:
Get the latest camera detections:
Stop the camera when finished:
7. Advanced Model Operations
Training a Custom Model
Validating a Model
Exporting a Model to Different Formats
8. Testing Connection
Check if the YOLO service is running correctly:
Troubleshooting
Camera Issues
If the camera doesn't work, try different camera IDs:
Model Not Found
If a model is not found, make sure you've downloaded it to one of the configured directories:
Performance Issues
For better performance with limited resources, use the smaller models (e.g., yolov8n.pt instead of yolov8x.pt)