Ncbi Datasets MCP server

@pipeworx/ncbi-datasets

NCBI Datasets (US National Library of Medicine) — which genome assemblies exist for an organism and how good they are, where a gene sits and what it is called in every other database, and the NCBI taxonomy tree with per-node assembly and gene counts.

Part of Pipeworx — an MCP gateway connecting AI agents to 1679+ live data sources.

Tools

— assemblies with level, length, contig/scaffold N50, GC%, submitter, BioProject and the paired RefSeq/GenBank accession.

• ncbigenomereports(taxon?, accession?, referenceonly?, assemblylevel?, limit?)

RefSeqGene coordinates, plus HGNC / Ensembl / UniProt / OMIM cross-references and every synonym.

• ncbigeneby_symbol(symbols, taxon?) — Entrez gene id, biotype, chromosome,

the full lineage, children, rank and live assembly/gene counts.

• ncbi_taxonomy(taxons) — resolve a tax id, scientific name or common name to

Auth

Keyless. NCBI rate-limits anonymous callers by IP at roughly 5 requests/second across api.ncbi.nlm.nih.gov; the pack sends an identifying User-Agent, does not retry, and surfaces a 429 with that explanation rather than looping.

Not the same service as ncbi-eutils

E-utilities is a generic search/fetch layer over ~40 Entrez databases that hands back records you must parse. Datasets answers structured questions about genomes, genes and taxonomy with structured rows. Use this pack for those three; use ncbi-eutils for PubMed, dbSNP and the rest of Entrez.

Data sources

Things that cost time to rediscover (measured 2026-09-17)

562 (E. coli) has 492,559, and page one of two is a GenBank/RefSeq pair of the same assembly — technically correct, practically useless. filters.referenceonly=true takes that to 2, so ncbigenome_reports defaults to it for a taxon lookup and reports that it did.

• Without filters, a genome report is every assembly ever deposited. Taxon

cross-linked by paired_accession. Counting rows double-counts assemblies.

• Every assembly exists twice, as GCA (GenBank) and GCF (RefSeq),

returns Mus musculus first. Match on the returned query / symbol, never on position.

• Multi-value paths do not preserve order. taxonomy/taxon/9606,10090

in taxonomy reports (tax_id: 9606). Both are passed through as received.

• Numeric fields are strings in gene reports ("gene_id":"672") but numbers

9606, Homo sapiens and human all work.

• A taxon path accepts a tax id, a scientific name or a common name —

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "ncbi-datasets": {
      "url": "https://gateway.pipeworx.io/ncbi-datasets/mcp"
    }
  }
}

What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/ncbi-datasets/mcp returns the tools in the table above plus the shared Pipeworx meta-tools — askpipeworx, discovertools, search_within, remember/recall and the rest of the gateway-wide set. So the tool count you see is larger than this table: a single-pack endpoint currently lists roughly 30 shared tools alongside the pack's own. The connection's initialize response states its exact scope, and is the authoritative answer for a given day.

Shortened. The full README is on GitHub.