Biothings MCP

What longevity-genie/biothings-mcp MCP server does

The longevity-genie/biothings-mcp MCP server connects MCP clients with BioThings data services. Its supported sources include mygene.info for gene annotations, myvariant.info for variant annotations, and mychem.info for chemical compound annotations. The repository also exposes taxonomic data through a dedicated taxon interface.

The project is intended for AI assistants and agents that need structured access to biomedical information instead of relying only on general language-model knowledge. It also includes Entrez download operations for retrieving data as returned content or saving it to a local file.

How it works

The server uses the biothings-typed-client library to provide typed client wrappers for the supported BioThings interfaces. The documented handlers are GeneTools, VariantTools, ChemTools, TaxonTools, and DownloadTools. MCP-compatible applications can discover and call these handlers after connecting through one of the available transports.

Three operating modes are documented. STDIO is the default and is intended for clients that communicate through standard input and output. Streamable HTTP starts a web server, using port 3001 by default, with the MCP endpoint at /mcp and API documentation at /docs. SSE provides a Server-Sent Events transport and also defaults to port 3001.

Setup and configuration

The longevity-genie/biothings-mcp MCP server can be run from PyPI with uvx:

uvx biothings-mcp

For explicit modes, the documented commands use uvx --from biothings-mcp stdio, uvx --from biothings-mcp server run, or uvx --from biothings-mcp sse. The HTTP and SSE commands accept host, port, and output-directory options where documented. A local checkout can instead be prepared with uv sync and run through uv run stdio, uv run server run, or uv run sse.

The repository includes configuration files for local and remote connections. It documents a hosted MCP endpoint at https://biothings.longevity-genie.info/mcp, so clients can connect remotely without running the server locally. Claude Desktop, Cursor, Windsurf, and other MCP-compatible clients are listed as integration targets.

Tools and capabilities

• Query gene annotation data through the gene interface.
• Query genetic variant annotation data through the variant interface.
• Query chemical compound data through the chemical interface.
• Query taxonomic information through the taxon interface.
• Download Entrez data as returned text through download_entrez_data.
• Download Entrez data to a local file through download_entrez_data_local.

Local downloads use biothings_output/ by default, unless --output-dir specifies another directory. The server creates the selected directory when needed and prefixes generated filenames with UUIDs to reduce filename collisions. Documented output extensions include FASTA, GenBank, alignment, JSON, and plain text formats.

Limitations and notes

The README describes the library as beta-quality. It identifies a problem with language models supplying invalid gene or variant symbols, and notes planned work on entity resolution and more descriptive method comments. Applications should therefore validate identifiers and inspect returned data rather than assuming that every natural-language entity reference will resolve correctly.

The repository documents tests for the API endpoint using uv run pytest -vvv -s. MCP Inspector can be used to inspect and test the available methods, but it is an optional debugging and exploration tool rather than a requirement for normal MCP-client usage.