Skip to main content
Glama
cellosaurus-mcp-server

cellosaurus-mcp-server

by QuentinCody

cellosaurus-mcp-server

MCP server for the SIB Cellosaurus API — the reference knowledge resource on cell lines (~169,000 cell lines: cancer lines, hybridomas, iPSC/ESC, patient-derived, etc.), each keyed by a stable CVCL_ accession and cross-referenced to ~100 other databases (CCLE/DepMap, Cosmic, ATCC, ChEMBL, PubMed, NCBI Taxonomy, NCIt disease, …).

  • Upstream API: https://api.cellosaurus.org (OpenAPI: https://api.cellosaurus.org/openapi.json)

  • Upstream site / docs: https://www.cellosaurus.org

  • Auth: none

  • License: CC BY 4.0 (surfaced as verifiable _meta.citation on every cellosaurus_execute result)

  • Response format: JSON by default (adapter sends Accept: application/json); format=xml|txt|tsv returned as { text }

  • API version: 1.0.4; dataset release is read at runtime from /release-info.

  • September 28, 2026 review observed release 56.0 (updated June 25), including new EDITGENE and NCI-PDMR cross-references.

Endpoints (curated from the upstream OpenAPI)

Only three GET methods exist upstream; all are curated in src/spec/catalog.ts:

  • GET /cell-line/{ac} — fetch one record by CVCL accession (category cell-line)

  • GET /search/cell-line — Solr search (q=, start, rows, fields, sort) (category search)

  • GET /release-info — dataset version / counts (category release)

GET /cell-line/{ac}/context is a local helper, composing the existing release and lookup APIs. It retains the native Cellosaurus envelope and adds _cellosaurus evidence.

REST only. Cellosaurus also serves SPARQL/RDF (https://sparql.cellosaurus.org) — noted as a future option, not wired here.

Related MCP server: mcp-uniprot

Tools (Code Mode only)

  • cellosaurus_search — discover the native endpoints, local context helper and recipes + usage notes from the catalog

  • cellosaurus_execute — run sandboxed JavaScript against the API via api.get() (carries _meta.citation)

  • cellosaurus_query_data — SQL over staged responses

  • cellosaurus_get_schema — inspect staged-dataset schemas

Large, well-annotated records (e.g. HeLa CVCL_0030 ≈ 450 KB, Caco-2 CVCL_0025 ≈ 130 KB compact JSON — verified live) auto-stage past the 30 KB inline threshold into CellosaurusDataDO; child arrays (comment-list, xref-list, reference-list, sequence-variation-list, …) become SQLite tables. Trim payloads with fields= (or a small rows=) to keep responses inline.

All four tools also register their mcp_cellosaurus_* aliases and carry caller-bound citations. Complete responses above 30 KiB stage, including composed output, SQL, schema and errors. payloads/content_chunks retain the full original response. Quote table names containing hyphens when querying nested native records.

Search pages add _cellosaurus request, source URL/hash, returned count and next_start_candidate. Upstream supplies no total or snapshot cursor; canonical completeness stays false, including empty or short pages. A next offset does not establish that another record exists. Searches may include publications alongside cell lines. Preserve both, along with each page’s source evidence.

Identity-aware joins

return await api.get('/cell-line/CVCL_0030/context', {
  xref_database: 'DepMap', xref_accession: 'ACH-001086',
  // Optional current-release guard, never a historical selector:
  expected_release: '56.0',
});

Use xref_database: 'PubChem_Cell_line' and its exact accession to check the PubChem link. Matching indexes point into the native record's xref-list; all matches survive. Empty matches describe only this record, not global absence. The helper requests acas so a former primary accession can resolve through a secondary accession. If native lookup returns 404, it performs a bounded exact acas: search and requires one record whose accession list contains the requested identifier. record_resolution records this fallback; ambiguity fails explicitly. Name matches, parent/child lines and lines from the same individual are not treated as equivalent models. Retain DepMap ModelID, ModelCondition and release, and PubChem assay/source metadata independently.

The helper preserves raw comments, STR allele strings, source/conflict records, and cell-line relationships. Warning indexes identify recognized problematic or caution comment categories; missing warnings do not establish authentication. No STR matching score or specimen-authentication result is computed.

Release metadata is checked before and after retrieval, with mismatches rejected. This detects observed drift, not an atomic snapshot. Record URL, observation time and SHA-256 of the original response bytes are retained separately from record entry dates/versions. Requests are bounded to 4 MiB each and a shared 50-second context deadline. Native requests have a 30-second deadline. Repeated fld parameters are supported and retain upstream precedence over fields.

Sources: API fields, release notes, HeLa identifiers.

October 3, 2026 native/local Worker acceptance passes through both aliases: full HeLa, context and exact DepMap/PubChem joins, an observed secondary accession, problematic-cell-line warnings, EDITGENE/NCI-PDMR pages, all four formats and errors. Recovered native payloads match independent reads. SQL preserves all 233 HeLa cross-references, 24 STR markers / 29 allele entries, 32 marker-source records, 17 profile-source records and 83 publications; all 27 generated relationships join. Deep marker sources remain JSON columns and match their native arrays exactly.

The final runs verify 3,284 MCP calls and citation proofs; complete 80 KB Unicode outputs and SQL results stage and recover. There are 84 server tests plus shared regressions. Production and historical snapshots remain unverified. Reproduction: scratch/2026-10-03-cellosaurus/ in the monorepo; acceptance details are in the implementation tracker.

Example (Code Mode)

// Resolve HeLa and list its cross-references (dr = xref-list)
const resp = await api.get('/cell-line/{ac}', { ac: 'CVCL_0030', fields: 'id,ac,ox,di,dr' });
if (resp.__staged) return resp; // Inspect its schema and query the staged records.
const cl = resp.Cellosaurus['cell-line-list'][0];
return {
  name: cl['name-list'].find(n => n.type === 'identifier')?.value,
  accession: cl['accession-list'].find(a => a.type === 'primary')?.value,
  xrefs: (cl['xref-list'] || []).map(x => `${x.database}:${x.accession}`),
};
// Find human cell lines that have a Cosmic-CLP cross-reference
const resp = await api.get('/search/cell-line', {
  q: 'ox:9606 AND dr:Cosmic-CLP', rows: 20, fields: 'id,ac,di',
});
if (resp.__staged) return resp;
return {accessions: resp.Cellosaurus['cell-line-list'].map(c => c['accession-list'][0].value),
  source: resp._cellosaurus};

License and data terms

The code in this repository is licensed under the Apache License 2.0 (see LICENSE). If you use it in your work, please cite it (see CITATION.cff). Citation is requested, not required.

The license covers the code only. Data that this server returns comes from Cellosaurus and stays under the terms of that source: CC BY 4.0. You are responsible for your use of the data under those terms.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers