t-neumann/GenomicOps-MCP

🧬 GenomicOps-MCP

A Model Context Protocol server for AI-assisted genomics workflows

Query UCSC genome browser tracks • Liftover coordinates between assemblies • Explore species and genomes

GenomicOps-MCP is a Model Context Protocol (MCP) server exposing genomic operations for AI-assisted bioinformatics workflows. It provides tools for querying UCSC genome browser tracks, performing coordinate liftover between assemblies, and exploring available species and assemblies — all accessible in dual mode:

• 🧠 Model Context Protocol (MCP) — for integration with Claude Desktop and other AI clients.
• 🌐 FastAPI REST API — for human-readable, local testing and web interaction.

It currently provides tools to query UCSC genome browser tracks, including feature overlaps and assembly listings.

🚀 Features

📁 Project Structure

GenomicOps-MCP/
├── server.py                 # FastMCP + FastAPI entrypoint
├── genomicops/
│   ├── __init__.py
│   ├── liftover.py           # Coordinate liftover logic
│   └── ucsc_rest.py          # UCSC REST API client
├── tests/
│   ├── conftest.py           # Pytest configuration
│   ├── test_api.py
│   ├── test_liftover.py
│   ├── test_mcp.py
│   └── test_ucsc_rest.py
├── pyproject.toml
└── uv.lock

🧩 Installation

1️⃣ Prerequisites

• Python ≥ 3.11
• uv (recommended for managing virtual environments and dependencies)

Install uv globally (if not already):

curl -LsSf https://astral.sh/uv/install.sh | sh

2️⃣ Clone and enter the project

git clone https://github.com/YOUR_USERNAME/GenomicOps-MCP.git
cd GenomicOps-MCP

3️⃣ Install dependencies via uv

uv sync

This ensures your environment matches the locked versions in uv.lock.

🧬 Available Tools

🧠 Running in MCP Mode

stdio Transport (for Claude Desktop)

Run the server as an MCP endpoint:

uv run server.py

This launches the MCP server over stdio, ready for a client like Claude Desktop to connect.

You should see something like:

────────────────────────────────────────────────────────────────────────────╮
│                                                                            │
│        _ __ ___  _____           __  __  _____________    ____    ____     │
│       _ __ ___ .'____/___ ______/ /_/  |/  / ____/ __ \  |___ \  / __ \    │
│      _ __ ___ / /_  / __ `/ ___/ __/ /|_/ / /   / /_/ /  ___/ / / / / /    │
│     _ __ ___ / __/ / /_/ (__  ) /_/ /  / / /___/ ____/  /  __/_/ /_/ /     │
│    _ __ ___ /_/    \____/____/\__/_/  /_/\____/_/      /_____(*)____/      │
│                                                                            │
│                                                                            │
│                                FastMCP  2.0                                │
│                                                                            │
│                                                                            │
│                 🖥️  Server name:     ucsc-mcp                               │
│                 📦 Transport:       STDIO                                  │
│                                                                            │
│                 🏎️  FastMCP version: 2.12.4                                 │
│                 🤝 MCP SDK version: 1.16.0                                 │
│                                                                            │
│                 📚 Docs:            https://gofastmcp.com                  │
│                 🚀 Deploy:          https://fastmcp.cloud                  │
│                                                                            │
╰────────────────────────────────────────────────────────────────────────────╯


[10/08/25 13:26:00] INFO     Starting MCP server 'ucsc-mcp' with transport 'stdio'

streamable-http Transport (for remote/web clients)

uv run fastmcp run server.py --transport streamable-http

Test with MCP Inspector:

npx @modelcontextprotocol/inspector --transport http --server-url http://localhost:8000/mcp

🧩 Add to Claude Desktop

To connect this MCP to Claude Desktop:

Option 1: Manual Configuration

Add this entry to your claude_desktop_config.json to run with default STDIO protocol:

{
  "mcpServers": {
    "GenomicOps-MCP": {
      "command": "uv",
      "args": [
        "run",
        "--project",
        "/path/to/GenomicOps-MCPP",
        "/path/to/GenomicOps-MCP/server.py"
      ]
    }
  }
}

Or install via FastMCP CLI:

fastmcp install claude-desktop server.py \
  --python 3.11 \
  --project /path/to/GenomicOps-MCP

🌐 Running in FastAPI Mode

For local testing with Swagger UI:

uv run server.py api

Open http://localhost:8000/docs for interactive API documentation.

REST Endpoints

Example Requests

# Get features overlapping MYC locus
curl -X POST http://localhost:8000/overlaps \
  -H "Content-Type: application/json" \
  -d '{"region": "chr8:127735433-127740477", "genome": "hg38", "track": "knownGene"}'

# Liftover coordinates from hg19 to hg38
curl -X POST http://localhost:8000/liftover \
  -H "Content-Type: application/json" \
  -d '{"region": "chr1:1000000-1001000", "from_asm": "hg19", "to_asm": "hg38"}'

# List human assemblies
curl http://localhost:8000/assemblies/Homo%20sapiens

🐳 Docker Deployment

Build

docker buildx build --platform linux/amd64,linux/arm64 -t tobneu/genomicops:latest --push .

Run Locally

docker run -d -p 8000:8000 --name genomicops tobneu/genomicops:latest

:cloud: AWS EC2 Deployment

1. Launch EC2 instance (Amazon Linux 2023 AMI, t3.small is sufficient)

   Configure security group:

   • SSH (port 22) from your IP
   • Custom TCP (port 8000) from your IP

2. Install Docker

   sudo yum update -y
   sudo yum install docker -y
   sudo systemctl start docker
   sudo systemctl enable docker
   sudo usermod -a -G docker ec2-user
   exit  # Log out and back in
   

3. Run container

   docker run -d -p 8000:8000 --name genomicops --restart unless-stopped tobneu/genomicops:latest
   

4. Verify

   npx @modelcontextprotocol/inspector --transport streamable-http --server-url http://l<EC2-IP>:8000/mcp
   

🧪 Testing

# Run unit tests
uv run pytest

# Run smoke tests (quick API validation)
uv run pytest --smoke

# Run integration tests (real UCSC API calls)
uv run pytest --integration

# Run all tests with coverage
uv run pytest --cov=genomicops --cov-report=html

⚠️ Limitations

• UCSC API Dependency: Requires network access to UCSC REST API
• Rate Limits: UCSC may rate-limit excessive requests
• Chain Files: liftOver requires downloading chain files (~1-50MB each) on first use
• Platform Support: liftOver binary auto-download supports Linux (x86_64) and macOS (x86_64)

🧰 Development Tips

• Run with uv

  uv run server.py
  

• Add new tools
  Decorate Python functions with @mcp.tool() inside server.py.
• Debug locally
  Use FastAPI mode (uv run server.py api) for quick JSON responses.

🐛 Troubleshooting

liftOver binary not found

The binary is auto-downloaded on first use. If it fails:

# Check platform detection
python -c "import platform; print(platform.system(), platform.machine())"

# Manual download (macOS example)
curl -O https://hgdownload.soe.ucsc.edu/admin/exe/macOSX.x86_64/liftOver
chmod +x liftOver
mv liftOver genomicops/liftover_data/

Chain file download fails

Chain files are fetched from UCSC. Ensure network access to hgdownload.soe.ucsc.edu.

UCSC API timeout

Increase timeout for slow connections:

curl "http://localhost:8000/tracks/hg38?timeout=30"

📚 References

• Model Context Protocol
• FastMCP Documentation
• UCSC Genome Browser REST API
• UCSC liftOver Tool

📄 License

This project is licensed under the MIT License - see the file for details.

Author: Tobias Neumann
Version: 0.1.0