🧬 Heuris-BioMCP — Bioinformatics Model Context Protocol Server

Strategic Model Context Protocol server for life sciences.
Connect ChatGPT, Claude, and other MCP clients to a curated biology tool surface built for research, translational workflows, and production review.

🚀 Quick Start • 🔧 Tools • 📊 Databases • 💡 Examples • 🤝 Contributing

Live Demo

Try Heuris-BioMCP without installing — connect to our live server:

https://heuris-biomcp.onrender.com/mcp

If hosted auth is enabled, Claude-compatible clients will complete the OAuth redirect flow automatically. Non-interactive clients can also use bearer API keys.

Connect to Live Server

For Claude, use Customize > Connectors and enter:

https://heuris-biomcp.onrender.com/mcp

For generic remote MCP clients that accept URL-based server definitions:

{
  "mcpServers": {
    "heuris-biomcp": {
      "url": "https://heuris-biomcp.onrender.com/mcp"
    }
  }
}

Note: https://heuris-biomcp.onrender.com/mcp is the recommended remote MCP endpoint for modern clients. The legacy SSE endpoint remains available at https://heuris-biomcp.onrender.com/sse.

Related MCP server: OrigeneMCP

See Heuris-BioMCP in Action

Quick Demo Video

Watch how to connect Heuris-BioMCP to Claude Desktop:

Tip: Coming soon - video walkthrough of connecting Heuris-BioMCP and running your first query!

What is Heuris-BioMCP?

Heuris-BioMCP bridges MCP clients and high-value life-science data sources through a curated public tool surface. The server exposes the workflows that matter most for product review and research use, while lower-level helper modules remain internal for composition and planner logic.

You -> "What drugs target EGFR and what clinical trials are recruiting?"
ChatGPT + Heuris-BioMCP -> Queries ChEMBL + ClinicalTrials.gov -> Structured answer

What's New in v2.3

• Curated the public MCP surface from a broad 71-tool registry down to a strategy-driven set of 32 review-friendly tools

• Merged operational suites into workflow tools: find_protein, pathway_analysis, crispr_analysis, drug_safety, variant_analysis, and session

• Added new translational tools: drug_interaction_checker, protein_binding_pocket, biomarker_panel_design, pharmacogenomics_report, protein_family_analysis, network_enrichment, rnaseq_deconvolution, structural_similarity, rare_disease_diagnosis, and genome_browser_snapshot

• Removed low-signal or niche tools from the public MCP registry while keeping lower-level code available internally

• Preserved hosted HTTP/SSE deployment and operational endpoints for production-style MCP review

• Added optional hosted OAuth 2.1 PKCE and API-key auth for authenticated remote connectors

• Added Prometheus metrics, streamed progress chunks for slow tools, persistent gene/disease/watch MCP resources, and literature watch workflows

Tools (32 curated public surface)

Core Research

AI And Engineering Workflows

High-Value Translational Tools

Public Surface Policy

• The MCP registry is intentionally curated.

• Lower-level legacy implementations still exist in the package for internal orchestration and testing.

• Reviewers should evaluate the exposed MCP surface, not the hidden implementation inventory.

Databases & AI Models

Quick Start

Option 1: Use Live Demo (No Installation)

Use Claude's Customize > Connectors flow and enter:

https://heuris-biomcp.onrender.com/mcp

If you are connecting with an older MCP client that still expects SSE, use:

https://heuris-biomcp.onrender.com/sse

Option 2: Deploy Your Own

Deploy to Render with one click:

Or manually:

1. Fork this repository

2. Create a new Web Service on Render

3. Connect your fork

4. Set build command: pip install -r requirements.txt && pip install -e .

5. Set start command: BIOMCP_TRANSPORT=http BIOMCP_HTTP_PORT=$PORT python -m biomcp

Hosted Deployment Limitations

• Session snapshots saved through the session tool are only durable if BIOMCP_SESSION_STORE_DIR points to persistent storage.

• On Render free tier, the default local directory uses ephemeral disk and will be wiped on restart, redeploy, or scale-to-zero wake-up.

• If you need persistent saved sessions, set BIOMCP_SESSION_STORE_DIR to a mounted persistent path or move session persistence behind an external store before relying on cross-session restore.

Hosted Auth and Connector Setup

• For Anthropic-style hosted connectors, enable BIOMCP_AUTH_ENABLED=1.

• OAuth 2.1 PKCE is exposed at /.well-known/oauth-authorization-server, /oauth/authorize, /oauth/token, and /oauth/register.

• For machine-to-machine clients, set BIOMCP_API_KEYS and send either Authorization: Bearer <key> or X-API-Key: <key>.

• Per-key limits are controlled with BIOMCP_API_KEY_RATE_LIMIT_REQUESTS and BIOMCP_API_KEY_RATE_LIMIT_WINDOW_SECONDS.

Persistent Resources and Literature Watches

• biomcp://gene/{SYMBOL} returns a curated gene context resource with gene, protein, pathway, disease, and drug-target context.

• biomcp://disease/{URL-ENCODED-NAME} returns disease literature plus session-graph context.

• session(action="watch") registers a PubMed + bioRxiv watch and exposes biomcp://watch/{TOPIC} as a reusable resource.

• Saved sessions and watches are only persistent if the backing session-store directory is durable.

Privacy, Support, and Data Handling

• Privacy policy: PRIVACY.md

• Support channel: SUPPORT.md

• Data handling notes: DATA_PROCESSING.md

• Security reporting: SECURITY.md

Option 3: Local Installation

Prerequisites

• Python 3.11+

• Claude Desktop or any MCP-compatible client

• (Optional) NCBI API key for higher rate limits

• (Optional) NVIDIA API keys for AI tools

Installation

# Clone the repository
git clone https://github.com/SachinGawande2003/Heuris-BioMCP.git
cd Heuris-BioMCP

# Install (standard)
pip install -e .

# Install with neuroimaging support
pip install -e ".[neuroimaging]"

# Install with dev dependencies (recommended)
pip install -e ".[dev]"

Configure Claude Desktop

For Local STDIO Mode (default):

{
  "mcpServers": {
    "heuris-biomcp": {
      "command": "biomcp",
      "env": {
        "NCBI_API_KEY": "your_ncbi_api_key_here"
      }
    }
  }
}

For Remote HTTP Mode (using live demo or your own deployed server):

Use Claude's Customize > Connectors flow and enter:

https://heuris-biomcp.onrender.com/mcp

If you are connecting with an older MCP client that still expects SSE, use:

https://heuris-biomcp.onrender.com/sse

💡 Tip: Get a free NCBI API key to increase rate limits from 3 to 10 requests/second.

🚀 New: Get free NVIDIA API keys for AI tools at build.nvidia.com/mit/boltz2 and build.nvidia.com/arc/evo2-40b.

Restart Claude Desktop and test:

"Search PubMed for recent papers on CAR-T cell therapy in B-cell lymphoma"
"Get the AlphaFold structure for TP53 and tell me about the confidence scores"
"What drugs are approved that target EGFR?"
"Generate a multi-omics report for KRAS"
"Predict the structure of EGFR with ligand CC1=CC=CC=C1 and compute binding affinity"
"Generate a DNA sequence starting with ATGGCG..."

Usage Examples

Literature Mining

"Search PubMed for BRCA1 CRISPR correction methods published in the last 2 years"
"Find review articles about PD-1/PD-L1 immune checkpoint inhibitors"

Protein Analysis

"Get UniProt info for human TP53 (P04637) including its domains and disease associations"
"Search for AlphaFold structures for insulin receptor"
"Find all PDB crystal structures of BRAF kinase domain resolved below 2.5 Ångström"

Drug Discovery

"What are the top ChEMBL compounds targeting KRAS G12C mutation?"
"Get compound info for imatinib (CHEMBL941)"
"Show me gene-disease associations for BRCA1 with evidence scores"

AI-Powered Structure Prediction

"Predict the 3D structure of insulin (sequence: ...) with ligand CCO"
"Compute binding affinity between EGFR and gefitinib (SMILES: ...)"
"Get structure prediction for a protein-protein complex"

AI-Powered DNA Generation

"Generate a 200bp promoter sequence starting with ATG"
"Compare wildtype vs variant DNA sequence for TP53 mutation"
"Generate regulatory element for gene expression"

Multi-Omics Report (Flagship)

"Generate a complete multi-omics report for EGFR"

This single command queries 7 databases in parallel and returns:

• Genomic location and gene summary (NCBI Gene)

• Recent publications (PubMed)

• Protein function and structure (UniProt + AlphaFold)

• Biological pathways (Reactome)

• Drug targets and clinical compounds (ChEMBL)

• Disease associations with scores (Open Targets)

• Expression datasets (GEO)

• Active clinical trials (ClinicalTrials.gov)

v2 Extended Databases

"Get OMIM diseases associated with TP53"
"Show STRING protein interactions for EGFR"
"Get GTEx expression data for BRCA1 across tissues"
"Find mutations in TP53 from cBioPortal"
"Search GWAS for diabetes-associated SNPs"

v2 Verification

"Verify the claim that TP53 is a tumor suppressor gene"
"Detect conflicts between OMIM and DisGeNET for BRCA1"

v2 Experimental Design

"Generate an experimental protocol for CRISPR knockout of BRCA1"
"What cell lines should I use to study KRAS mutations?"
"Calculate sample size for detecting 2-fold change with p<0.05"

v2 Session Intelligence

"What's the knowledge graph from our conversation so far?"
"Find biological connections between TP53 and EGFR"
"Export our research session as a reproducible script"
"Plan and execute a research workflow for PD-1 drug targets"

Architecture

biomcp/
├── src/biomcp/
│   ├── server.py              # MCP server — tool registry & dispatcher
│   ├── tools/
│   │   ├── ncbi.py            # PubMed, Gene, BLAST
│   │   ├── proteins.py        # UniProt, AlphaFold, PDB
│   │   ├── pathways.py        # KEGG, Reactome, ChEMBL, Open Targets
│   │   ├── advanced.py        # ClinicalTrials, GEO, scRNA, Ensembl,
│   │   │                      # Multi-Omics, Neuroimaging, Hypothesis
│   │   ├── nvidia_nim.py      # Boltz-2, Evo2-40B AI models
│   │   ├── databases.py       # v2: OMIM, STRING, GTEx, cBioPortal,
│   │   │                      #      GWAS, DisGeNET, PharmGKB
│   │   ├── verify.py          # v2: Claim verification, conflict detection
│   │   └── protocol_generator.py  # v2: Experimental design tools
│   ├── core/
│   │   ├── entity_resolver.py # v2: Cross-database entity resolution
│   │   ├── knowledge_graph.py # v2: Session knowledge graph
│   │   └── query_planner.py   # v2: Adaptive query planner
│   └── utils/
│       └── __init__.py        # Rate limiter, cache, validators, HTTP client
├── tests/
├── pyproject.toml
└── README.md

Key design decisions:

• Async-first: All API calls are fully async with httpx, never blocking

• Rate limiting: Token-bucket limiter per service respects each API's limits

• Smart caching: TTL-based per-namespace cache (1h literature, 7d structures)

• Retry logic: Exponential backoff via tenacity for transient failures

• Validation: Input validation before any network call — never wastes API quota

Environment Variables

Get free API keys:

• NCBI: https://www.ncbi.nlm.nih.gov/account/

• NVIDIA Boltz-2: https://build.nvidia.com/mit/boltz2

• NVIDIA Evo2-40B: https://build.nvidia.com/arc/evo2-40b

• BioGRID: https://webservice.thebiogrid.org/

Operational Endpoints

When BioMCP runs in hosted HTTP mode, these operational routes are available:

These endpoints are designed for deployment review, Render health checks, and production smoke tests.

Contributing

Contributions are welcome! Whether it's adding a new database, fixing a bug, improving documentation, or integrating new AI models.

# Development setup
pip install -e ".[dev]"

# Run tests
pytest tests/ -v --cov=biomcp

# Lint + type check
ruff check src/
mypy src/

Ideas for contributions

• Add more pathway databases (Wikipathways, PathCards)

• Integrate COSMIC for somatic mutations

• Add protein complex data (CORUM)

• Implement batch query support for high-throughput analysis

• Add Jupyter notebook examples

• Improve conflict resolution algorithms

Citation

If you use Heuris-BioMCP in your research, please cite:

@software{biomcp2025,
  title   = {Heuris-BioMCP v2: A Comprehensive MCP Server for Bioinformatics, AI Models, and Life Sciences},
  year    = {2025},
  url     = {https://github.com/SachinGawande2003/Heuris-BioMCP},
  license = {MIT}
}

License

MIT License — free for academic and commercial use.