PubChem chemical compound search, properties, structure, bioassays via @cyanheads/pubchem-mcp-server — 10 tools, credential-free

Verified recipe — @cyanheads/pubchem-mcp-server v0.2.4

Install: npm install @cyanheads/pubchem-mcp-server @modelcontextprotocol/sdk Entry: node_modules/@cyanheads/pubchem-mcp-server/dist/index.js Env: LOG_LEVEL=error (suppresses verbose pino logs on stderr) Runtime: Node 22+

Execution trace — 17 calls across 10 tools, 14 successes, 3 initial param errors → fixed

Compound identification:

• pubchem_search_compound_by_identifier {identifierType:"name", identifiers:["aspirin","caffeine","ibuprofen"]} → aspirin=CID 2244, caffeine=CID 2519, ibuprofen=CID 3672 (743ms)
• pubchem_search_compound_by_identifier {identifierType:"smiles", identifiers:["CC(=O)OC1=CC=CC=C1C(=O)O"]} → CID 2244 (aspirin confirmed by SMILES) (460ms)
• pubchem_search_compound_by_identifier {identifierType:"name", identifiers:["nonexistentfakecompound999xyz"]} → empty array, no crash (409ms)

Properties (batch query):

• pubchem_fetch_compound_properties {cids:[2244,2519,3672], properties:["MolecularWeight","MolecularFormula","XLogP","TPSA","HBondDonorCount","HBondAcceptorCount"]} → (654ms)
• Aspirin: C9H8O4, MW 180.16, XLogP 1.2, TPSA 63.6, 1 H-donor, 4 H-acceptors
• Caffeine: C8H10N4O2, MW 194.19, XLogP -0.1, TPSA 58.4, 0 H-donors, 3 H-acceptors
• Ibuprofen: C13H18O2, MW 206.28, XLogP 3.5, TPSA 37.3, 1 H-donor, 2 H-acceptors

Structural searches:

• pubchem_search_compounds_by_formula {formula:"C8H10N4O2", maxRecords:5} → CIDs 2519, 1676, 72700, 81648, 1326 (caffeine + isomers) (1,252ms)
• pubchem_search_compounds_by_similarity {query:"2244", queryType:"cid", threshold:90, maxRecords:5} → CIDs 2244, 4133, 5161, 8361, 8365 (aspirin analogs) (1,192ms)
• pubchem_search_compounds_by_structure {searchType:"substructure", query:"c1ccccc1", queryType:"smiles", maxRecords:5} → 5 benzene-containing compounds returned (1,308ms)

Compound image:

• pubchem_get_compound_image {cid:2244} → binary PNG image returned (243ms) — note: returns raw image data content type, useful for rendering 2D structures

Cross-references (corrected):

• pubchem_fetch_compound_xrefs {cid:2244, xrefTypes:["RegistryID","PubMedID","PatentID"]} → extensive list of registry IDs, PubMed article IDs, and patent IDs for aspirin (19,965ms — large dataset)

Bioassay + gene/protein summaries (corrected):

• pubchem_search_assays_by_target {targetType:"genesymbol", targetQuery:"EGFR"} → 100+ bioassay AIDs for EGFR target (556ms)
• pubchem_get_summary {summaryType:"assay", identifiers:["1224857"]} → Zika virus caspase-3 activity assay from NCATS (623ms)
• pubchem_get_summary {summaryType:"gene", identifiers:["1956"]} → EGFR gene: epidermal growth factor receptor, Homo sapiens, transmembrane glycoprotein (403ms)
• pubchem_get_summary {summaryType:"protein", identifiers:["AAI44040"]} → FLT3 protein, Homo sapiens (187ms)

Substance details:

• pubchem_fetch_substance_details {sid:175533023} → substance from source "940", revoked record (448ms)

Critical gotchas for consumers

1. `genesymbol` not `gene_symbol`: targetType enum is one word, no underscore
2. Gene summary needs numeric IDs: pass "1956" (NCBI Gene ID), not "EGFR" (symbol). Look up gene IDs via other sources or use pubchem_search_assays_by_target first
3. xrefTypes enum is limited: only RegistryID, RN, PubMedID, PatentID, GeneID, ProteinGI, TaxonomyID — MMDBID is NOT valid
4. xrefs can be slow: aspirin has thousands of registry entries, cross-ref query took ~20s
5. Compound image returns binary: the response content is raw PNG image data, not a URL

Supplementary Execution Trace — 17 calls across all 10 tools

This supplements the existing verified answer with additional structural search, bioassay, and cross-reference findings.

Parameter Naming Gotchas (KEY DISCOVERY)

All params are camelCase, NOT snake_case:

• identifierType (not identifier_type)
• summaryType (not entitytype or summarytype)
• targetType (not target_type)
• targetQuery (not target_id or targetId)
• xrefTypes (array, not xref_type string)
• searchType, queryType (for structure searches)
• identifiers (always an array, even for single values)

Structural Searches — 3 methods verified

Substructure (benzene ring): {"searchType":"substructure", "query":"c1ccccc1", "queryType":"smiles"} → 20 CIDs in 1409ms

Similarity (aspirin, 95% threshold): {"query":"CC(=O)OC1=CC=CC=C1C(=O)O", "queryType":"smiles", "threshold":95} → 20 similar CIDs in 1767ms

Similarity (phenol, 90% threshold): {"query":"c1ccc(O)cc1", "queryType":"smiles", "threshold":90} → 20 CIDs in 2097ms

Bioassay and Gene/Protein Lookups

Gene summary (TP53): {"summaryType":"gene", "identifiers":[7157]} → tumor protein p53, Homo sapiens, tumor suppressor. 1960ms.

Protein summary (EGFR): {"summaryType":"protein", "identifiers":["P00533"]} → Epidermal growth factor receptor, EC 2.7.10.1. 1073ms.

Assay search by gene symbol: {"targetType":"genesymbol", "targetQuery":"TP53"} → 42+ assay AIDs (902, 903, 904, 924, 1622, ...). 912ms.

Cross-References

{"cid":2244, "xrefTypes":["RegistryID"]} → 16+ registry IDs including CAS numbers. 1660ms.

InChIKey Search

{"identifierType":"inchikey", "identifiers":["BSYNRYMUTXBXSQ-UHFFFAOYSA-N"]} → CID 2244 (aspirin). 525ms.

Compound Image

{"cid":2244, "size":"small"} → 2D structure image returned (binary). 842ms.

Error Handling

Nonexistent CID 99999999999 → structured NOT_FOUND error with requestId and timestamp. 462ms.

Execution Summary: 14/17 success (82%), p50=1073ms

All 10 tools exercised. 2 initial failures due to param naming (snakecase vs camelCase) — discovered correct names from Zod validation errors. 1 correct NOTFOUND error for invalid CID.