BioMCP: Biomedical Model Context Protocol Server¶

BioMCP is a specialized Model Context Protocol (MCP) server that connects AI assistants like Claude to biomedical data sources, including ClinicalTrials.gov, PubMed, MyVariant.info, and cBioPortal.

Built and Maintained by ¶

Quick Start: Claude Desktop Setup¶

The fastest way to get started with BioMCP is to set it up with Claude Desktop:

Install Claude Desktop from Anthropic
Ensure uv is installed:

# Install uv if you don't have it
# MacOS: brew install uv
# Windows: pip install uv

Configure Claude Desktop:
Open Claude Desktop settings
Navigate to Developer section
Click "Edit Config" and add:

{
  "mcpServers": {
    "biomcp": {
      "command": "uv",
      "args": ["run", "--with", "biomcp-python", "biomcp", "run"]
    }
  }
}

Save and restart Claude Desktop
Start chatting with Claude about biomedical topics!

For detailed setup instructions and examples, see our Claude Desktop Tutorial.

What is BioMCP?¶

BioMCP is a specialized MCP (Model Context Protocol) server that bridges the gap between AI systems and critical biomedical data sources. While Large Language Models (LLMs) like Claude have extensive general knowledge, they often lack real-time access to specialized databases needed for in-depth biomedical research.

Using the Model Context Protocol, BioMCP provides Claude and other AI assistants with structured, real-time access to:

Clinical Trials - Searchable access to ClinicalTrials.gov for finding relevant studies
Research Literature - Query PubMed/PubTator3 for the latest biomedical research with automatic cBioPortal cancer genomics integration
Genomic Variants - Explore detailed genetic variant information from MyVariant.info
Cancer Genomics - Automatic integration with cBioPortal for mutation occurrence data in cancer studies

Through MCP, AI assistants can seamlessly invoke BioMCP tools during conversations, retrieving precise biomedical information without the user needing to understand complex query syntax or database-specific parameters.

MCP Tools and Capabilities¶

BioMCP provides 13 specialized tools through the MCP interface:

Core Tools (3)¶

Think Tool (CRITICAL - ALWAYS USE FIRST!)¶

think: A sequential thinking tool for systematic analysis of biomedical problems. MUST be used BEFORE any search operations to ensure comprehensive research planning and analysis.

Unified Tools¶

search: Powerful unified search across all biomedical data sources with cross-domain query support
fetch: Retrieve detailed information for specific articles, trials, or variants

Individual Tools (10)¶

Article Tools (2)¶

article_searcher: Search PubMed/PubTator3 and preprints with automatic cBioPortal integration
article_getter: Fetch detailed article content and metadata

Note: When searching articles with gene parameters, cBioPortal data is automatically included, providing:

Gene-level mutation summaries across cancer studies
Mutation-specific search capabilities (e.g., BRAF V600E)
Dynamic cancer type categorization

Clinical Trial Tools (5)¶

trial_searcher: Search for trials by condition, intervention, location, phase, etc.
trial_getter: Fetch all details for a specific trial
trial_protocol_getter: Get protocol information only
trial_references_getter: Find publications related to trials
trial_outcomes_getter: Access trial results and outcome data
trial_locations_getter: Find where trials are conducted

Genomic Variant Tools (2)¶

variant_searcher: Search MyVariant.info with clinical and functional filters
variant_getter: Get comprehensive annotations including TCGA, 1000 Genomes, and cBioPortal data

MCP Resources¶

BioMCP provides the following resources through the MCP interface:

Instructions Resource¶

get_instructions: Returns operational instructions and guidelines for effective use of BioMCP tools. This resource helps AI assistants understand best practices for biomedical research tasks.

Researcher Persona Resource¶

get_researcher: Provides a detailed biomedical researcher persona with extensive expertise across multiple domains. This persona helps AI assistants adopt appropriate research methodologies and communication styles for biomedical tasks.

Tutorials¶

Getting Started¶

Claude Desktop Tutorial - Set up and use BioMCP with Claude Desktop
MCP Inspector Tutorial - Test and debug BioMCP directly
Python SDK Tutorial - Use BioMCP as a Python library
MCP Client Tutorial - Integrate with MCP clients programmatically

Advanced Features¶

AlphaGenome Setup Guide - Configure Google DeepMind's AlphaGenome for variant effect prediction
AlphaGenome Prompt Examples - Example prompts and workflows for variant analysis with AI
AlphaGenome with Docker - Run AlphaGenome in Docker containers

Verification and Testing¶

The easiest way to test your BioMCP setup is with the MCP Inspector:

npx @modelcontextprotocol/inspector uv run --with biomcp-python biomcp run

This launches a web interface where you can test each BioMCP tool directly. For detailed instructions, see the MCP Inspector Tutorial.

Additional Usage Options¶

While BioMCP is primarily designed as an MCP server for AI assistants, it can also be used in other ways:

Command Line Interface¶

BioMCP includes a comprehensive CLI for direct interaction with biomedical databases:

# Examples of CLI usage
biomcp trial search --condition "Melanoma" --phase PHASE3
biomcp article search --gene BRAF --disease Melanoma  # Includes cBioPortal data
biomcp article search --gene BRAF --keyword V600E     # Mutation-specific search
biomcp variant search --gene TP53 --significance pathogenic

Python SDK¶

For programmatic access, BioMCP can be used as a Python library:

# Install the package
pip install biomcp-python

See the Python SDK Tutorial for code examples.

MCP Client Integration¶

For developers building MCP-compatible applications, BioMCP can be integrated using the MCP client libraries. See the MCP Client Tutorial for details.

License¶

BioMCP is licensed under the MIT License.