MCP Server Cvdlt

🚀 YOLO Server Based on MCP (Computer Vision and Deep Learning Tool)

This repository is built upon the Model Context Protocol of Ultralytics and the Python SDK. It provides a server implementation that supports image object detection, segmentation, and human pose estimation.

Relevant Links

• MCP Playground Client - https://github.com/MRonaldo-gif/mcp-playground-local
• Ultralytics - https://github.com/ultralytics/ultralytics
• MCP Python Version - https://github.com/modelcontextprotocol/python-sdk

The server implements the Model Context Protocol in Python, offering functionalities for image object detection, segmentation, and human pose estimation.

✨ Features

• Detect objects in images using YOLOv10.
• Segment objects in images using YOLOv8.
• Segment the entire image using Ultralytics SAM.
• Estimate human poses in images using YOLOv8.
• Support both local and network image inputs.
• Integrate MCP tools for client interaction.
• Support Stdio and SSE transmission protocols.

⚠️ Important Note

The server requires valid image paths or URLs and access to the following model files: yolov10b.pt (YOLOv10 detection), yolov8n-seg.pt (YOLOv8 segmentation), yolov8n-pose.pt (YOLOv8 pose estimation), and sam_b.pt (Ultralytics SAM).

📋 To-Do List

• 3D detection
• AIGC (Generative Adversarial Networks, Diffusion Models)
• Density estimation
• Deployment of deep learning models

🚀 Quick Start

📦 Installation

uv sync  
# If you need the Tsinghua University mirror
uv sync --index https://pypi.tuna.tsinghua.edu.cn/simple --extra-index-url https://pypi.org/simple  

uv pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

💻 Starting the Server

1. Stdio Mode:

   python server.py
   

   Output:

   Starting MCP Server (YOLO) with stdio transmission
   

2. SSE Mode:

   python server.py sse [port number]
   

   Example:

   python server.py sse 8080
   

   Output:

   Starting MCP Server (YOLO) on port 8080 with SSE transmission
   

In addition, users need to download the weight files to the ./checkpoints directory. Download Link 🔗: https://...

💻 Using Claude Desktop

Add the following content to your claude_desktop_config.json:

⚠️ Important Note

You can provide a read-only directory /projects to the server by mounting a sandbox directory.

SSE Mode Configuration

{
  "mcpServers": {
    "server-with-yolo": {
      "url": "http://localhost:8080/sse"
    }
  }
}

Weight File Directory Structure:

./checkpoints/  
├── yolov10b.pt  
├── yolov8n-seg.pt  
├── yolov8n-pose.pt  
└── sam_b.pt