/projects/genomemcp
A research-grade MCP server that lets AI agents query clinical genomics in real time — ClinVar variant interpretation, gnomAD population frequencies, Reactome pathways and PubMed evidence — with a rich CLI and interactive TUI.
Genomic agents need access to trusted biomedical sources, but most assistant integrations lack a protocol-safe way to query ClinVar, gnomAD, Reactome, PubMed, and gene context in real time.
GenomeMCP exposes clinical genomics tools over MCP with CLI and TUI surfaces, letting agents query variants, genes, literature, population stats, pathways, genomic context, and discovery evidence.
The system must normalize external source responses, keep tool contracts clear for agents, distinguish evidence types, and make biomedical retrieval useful without pretending it is a diagnosis engine.
The project converts genomics retrieval into a protocol-native agent interface. MCP becomes the boundary that lets different clients call the same biomedical tools safely and consistently.
This highlights Stefan's rare intersection of MCP server design, bioinformatics retrieval, clinical genomics source literacy, Python CLI/TUI tooling, and agent-ready tool contracts.
GenomeMCP is a protocol-native bridge between agents and trusted genomics sources. It turns biomedical retrieval into callable tools with clear contracts instead of one-off search behavior.
ClinVar, gnomAD, Reactome, PubMed, and coordinate-context tools each provide different evidence. The MCP server normalizes them without flattening their meaning.
Agent clients need stable names, arguments, and response shapes. MCP gives the assistant a precise way to ask for variant reports, gene info, literature, population frequency, and pathways.
Clinical significance, population frequency, pathway context, and literature support are not interchangeable. The system keeps evidence types visible so agents can reason more carefully.
CLI and TUI access make the same tools useful outside an MCP client. That matters for researchers who want scriptable retrieval and humans who want quick inspection.
A practical fit for genomics clients, CLI work, and fast integration with scientific APIs.
Lets Claude Desktop, Cursor, Windsurf, and other agents call the genomics tools through one protocol.
Provides a clean command-line surface for researchers who do not want to open an MCP client.
Makes local stdio runs and hosted startup repeatable with a simple Python toolchain.
Searches ClinVar for genes, variants, or diseases.
Returns clinical significance and interpretation details for a variant.
Fetches gene function, location, aliases, and annotations.
Pulls PubMed evidence that supports an interpretation.
Queries gnomAD for allele frequency and population context.
Looks up Reactome pathway data for a gene.
Turns pathway relationships into a Mermaid diagram.
Finds genes linked to a phenotype or disease area.
Maps coordinates into exon, intron, and nearby gene context.
Aggregates abstracts and database signals for exploratory agent reasoning.