How do I use MedVision MCP?

1. Click on "Install Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@MedVision MCP Analyze this chest X-ray and describe any clinical findings." That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

MedVision MCP

Medical Vision AI Tools via Model Context Protocol (MCP)

Overview

MedVision MCP provides AI-powered medical image analysis tools accessible through the Model Context Protocol. It enables LLM agents (like Claude, GitHub Copilot) to analyze chest X-rays using Visual RAG (RAD-DINO + FAISS + DenseNet).

Features

✅ DenseNet Classification: 18 pathology detection (Lung Opacity, Pneumonia, etc.)
✅ RAD-DINO Embeddings: 768-dim visual embeddings for similarity search
✅ FAISS Index: Fast similarity search for similar historical cases
✅ DICOM Support: Native DICOM file reading
✅ Gradio Canvas: Interactive ROI drawing/annotation interface
✅ ROI Analysis: Analyze specific regions drawn on X-rays
🔜 Medical SAM: SAM-based region segmentation

Quick Start

# Clone git clone https://github.com/u9401066/medvision-mcp.git cd medvision-mcp # Install with uv uv sync # Test classification uv run python -c " import asyncio from src.medvision_mcp.server import classify_xray async def main(): result = await classify_xray('path/to/xray.dcm') print(result) asyncio.run(main()) "

MCP Tools

Tool	Description
`analyze_xray`	Full Visual RAG analysis (classification + similarity)
`classify_xray`	Quick DenseNet-121 classification (18 pathologies)
`search_similar_cases`	RAG similarity search
`build_rag_index`	Build FAISS index from image directory
`load_rag_index`	Load pre-built index
`get_engine_status`	Check model loading status

Gradio UI

Launch the interactive web UI:

# Start Gradio server uv run python -m src.medvision_mcp.ui.app # Open http://localhost:7860

UI Tabs:

Tab	Description
📊 Analysis	Full image analysis (classification + RAG)
⚡ Quick Classify	Fast 18-pathology classification
🎨 Canvas ROI	Draw ROIs and analyze specific regions
🔧 Build Index	Create FAISS index from images
📂 Load Index	Load pre-built index
ℹ️ Status	Check model loading status

Claude Desktop Configuration

Add to ~/.config/claude/claude_desktop_config.json:

{ "mcpServers": { "medvision": { "command": "uv", "args": ["run", "--directory", "/path/to/medvision-mcp", "python", "-m", "src.medvision_mcp.server"] } } }

Architecture

┌─────────────────────────────────────────────────────────┐ │ MCP Client (Claude, Copilot) │ └─────────────────────────┬───────────────────────────────┘ │ stdio ┌─────────────────────────▼───────────────────────────────┐ │ MedVision MCP Server │ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────────┐ │ │ │ classify │ │ search │ │ analyze │ │ │ │ _xray │ │ _similar │ │ _xray │ │ │ └─────────────┘ └─────────────┘ └─────────────────┘ │ │ │ │ │ ┌───────────────────────▼────────────────────────────┐ │ │ │ Visual RAG Engine │ │ │ │ RAD-DINO │ FAISS │ DenseNet-121 │ │ │ └────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────┘

Development

# Install dev dependencies uv sync --dev # Run tests uv run pytest # Check types uv run pyright

License

MIT

MedVision MCP

MedVision MCP

Overview

Features

Quick Start

MCP Tools

Gradio UI

Claude Desktop Configuration

Architecture

Development

License

Resources

Latest Blog Posts

MCP directory API