VCF MCP Server: Exposing Variant Calling Format files to LLMs for Analysis

No warranty express or implied. This was totally vibe-coded while chronicly sleep-deprived and watching Bluey and Tremors 5. Use at your own risk. AFAICT it works and outputs line up with what I get through traditional tools.

Overview

This is an MCP server that exposes a VCF (Variant Calling Format) file to LLMs for genomic variant analysis. The server provides tools to query variants by genomic position, region, or variant ID.

It is not intended to serve up multiple VCF files at once. It is designed to be used with a single VCF file in a desktop or confidential compute VM setting.

Prerequisites

• Rust 1.70 or later

Build

You want to build in release mode for optimal performance:

cargo build --release

The binary will be at ./target/release/vcf_mcp_server

Usage

• stdio transport: ./target/release/vcf_mcp_server sample_data/sample.compressed.vcf.gz
• HTTP/SSE transport: ./target/release/vcf_mcp_server sample_data/sample.compressed.vcf.gz --sse 127.0.0.1:8090

Options

• --sse <ADDR:PORT> - Run HTTP server on specified address (e.g., 127.0.0.1:8090)
• --debug - Enable debug logging
• --never-save-index - Never save the built tabix index to disk (for read-only/ephemeral environments)

Available MCP Tools

1. query_by_position

Query variants at a specific genomic position.

Parameters:

• chromosome (string): Chromosome name (e.g., '1', '2', 'X', 'chr1')
• position (integer): Genomic position (1-based)

Example:

{
  "name": "query_by_position",
  "arguments": {
    "chromosome": "20",
    "position": 14370
  }
}

2. query_by_region

Query variants in a genomic region.

Parameters:

• chromosome (string): Chromosome name (e.g., '1', '2', 'X', 'chr1')
• start (integer): Start position (1-based, inclusive)
• end (integer): End position (1-based, inclusive)

Example:

{
  "name": "query_by_region",
  "arguments": {
    "chromosome": "20",
    "start": 14000,
    "end": 18000
  }
}

3. query_by_id

Query variants by variant ID (e.g., rsID).

Parameters:

• id (string): Variant ID (e.g., 'rs6054257')

Example:

{
  "name": "query_by_id",
  "arguments": {
    "id": "rs6054257"
  }
}

VCF File Requirements

Compressed VCF Files (Recommended)

For efficient querying by position/region, use bgzip-compressed VCF files with tabix indexes. You probably already have these files if you're using a genome browser or other tools. If not, you can create them using the following commands:

1. Compress with bgzip:

   bgzip myfile.vcf
   # Creates myfile.vcf.gz

2. Create tabix index:

   tabix -p vcf myfile.vcf.gz
   # Creates myfile.vcf.gz.tbi

The server will automatically use the .tbi index file if present, or build an in-memory index. The tabix file will be saved it alongside your VCF file if it doesn't already exist and --never-save-index was not used.

Uncompressed VCF Files

Uncompressed VCF files are supported but will be indexed in-memory only. For large files, this can be slow and memory-intensive.

Development

The usual for Rust development:

cargo run
cargo test
cargo fmt
cargo clippy
cargo bench  # Run performance benchmarks

Test Data

Sample VCF files in the sample_data/ directory are provided for testing and demonstration:

• VCFlib samples - Small test files from the VCFlib project (MIT License)
• 1000 Genomes Project data - Subset of high-coverage sequencing data (CC BY-NC-SA 3.0)

See sample_data/README.md for complete attribution and licensing information.

License

MIT