Search PubMed for medical/scientific articles and fetch abstracts via mcp-simple-pubmed (uvx)

Supplementary verification: 13 additional calls, 100% success (run 2026-06-20)

Fresh execution of mcp-simple-pubmed v0.1.16 with 13 tool calls across both tools, verifying capabilities not covered in the original recipe.

New findings

1. MeSH term queries work perfectly:

search_pubmed({query: '"artificial intelligence"[MeSH] AND "drug discovery"[MeSH]', max_results: 2})
→ 2 results, structured JSON with pmid, title, abstract, journal, authors, DOI

Latency: 1156ms

2. Complex boolean operators + date filtering:

search_pubmed({query: '(SARS-CoV-2 OR COVID-19) AND vaccine AND mRNA AND 2025[pdat]', max_results: 2})
→ 2 results with papers from 2025 publication dates

Latency: 1899ms. Parenthesized OR, multi-term AND, and date range all work in a single query.

3. Author + topic filtering:

search_pubmed({query: 'Smith J[Author] AND diabetes', max_results: 2})
→ 2 results from authors named "Smith J" on diabetes-related papers

Latency: 1833ms

4. Full-text XML retrieval for PMC articles CONFIRMED:

get_paper_fulltext({pmid: "33856027"})  → JATS XML, 1821ms (PMC ID 8083258 found automatically)
get_paper_fulltext({pmid: "30271870"})  → JATS XML, 972ms  (PMC ID 6160392 found automatically)

Returns raw JATS/NLM XML with full document structure (front-matter, body, references, tables, figures). The server automatically resolves PMID → PMC ID via NCBI's id converter API.

5. Non-PMC articles handled gracefully:

get_paper_fulltext({pmid: "32817011"})
→ "Full text is not available in PubMed Central.\n\nThe article may be available at these locations:\n- PubMed page: https://pubmed.ncbi.nlm.nih.gov/32817011/\n- Publisher's site (via DOI): https://doi.org/10.1136/bmjebm-2020-111336"

Provides alternative URLs instead of crashing.

6. Invalid PMID handled gracefully:

get_paper_fulltext({pmid: "9999999999"})
→ same "not available" response with PubMed URL (no crash)

7. Empty result set:

search_pubmed({query: "xyznonexistenttopic12345", max_results: 5})
→ [] (empty array, no error)

Latency: 527ms

Key gotchas not in original recipe

• Full text is JATS XML — raw <?xml ...> with NLM DTD, not plain text or Markdown. Intentional per author (preserves structure for AI parsing).
• PMC lookup adds ~500ms — server calls NCBI id converter before fetching text
• `max_results` reliably caps results — tested 1, 2, 3, 5; all return exact count or fewer
• 3 MCP Prompts available but not exercised: systematic_review_search, pico_search, author_search — for constructing optimized PubMed queries

Performance across 13 calls

• p50: 1200ms (network-bound to NCBI servers)
• All 13/13 success
• Search: ~1000-2300ms depending on result count
• Fulltext PMC: ~1000-1800ms
• Empty search: ~500ms

Recipe: Search PubMed for biomedical articles via mcp-simple-pubmed (uvx)

Package: mcp-simple-pubmed v0.1.16 on PyPI Transport: stdio (via FastMCP 2.14.1) Auth: PUBMED_EMAIL env var required — any email address (NCBI's courtesy rate-limit field, not a credential) Launch: PUBMED_EMAIL="[email protected]" uvx mcp-simple-pubmed

Tools available (2)

Search syntax

• Simple keywords: "covid vaccine"
• Field-specific: "breast cancer"[Title], "Smith J"[Author], "RNA"[MeSH Terms]
• Date ranges: "2024:2026"[Date - Publication]
• Boolean: combine with AND, OR, NOT

MCP handshake trace (real execution 2026-06-12)

Step 1 — initialize:

→ {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"pathfinder","version":"1.0"}}}
← {"jsonrpc":"2.0","id":1,"result":{"protocolVersion":"2024-11-05","capabilities":{"tools":{"listChanged":true},"prompts":{"listChanged":false},"resources":{"subscribe":false,"listChanged":false}},"serverInfo":{"name":"pubmed-server","version":"2.14.1"}}}

Step 2 — tools/call (search for CRISPR gene therapy):

→ {"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search_pubmed","arguments":{"query":"CRISPR gene therapy 2024","max_results":3}}}
← (3 results returned, each with full structured data)

Sample result (trimmed):

{
  "pmid": "42080267",
  "title": "AlphaFold3-guided tracrRNA redesign yields small monomeric Cas12f RNPs.",
  "journal": "Nucleic acids research",
  "authors": ["Pan Lulu", "Sang Rui", "Xue Ruier", "Ma Yongcheng", "Goldys Ewa", "Deng Fei"],
  "publication_date": {"year": "2026", "month": "Apr", "day": "23"},
  "doi": "10.1093/nar/gkag430",
  "pmcid": "PMC13137049",
  "pubmed_url": "https://pubmed.ncbi.nlm.nih.gov/42080267/",
  "pmc_url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13137049/",
  "abstract": "Although Cas12f (Cas14) is among the smallest Class 2 CRISPR effectors...",
  "mesh_terms": [{"descriptor": "CRISPR-Cas Systems", "ui": "D064113"}, ...],
  "keywords": []
}

Notes

• Returns richly structured JSON: pmid, title, authors, abstract, journal, publication_date, DOI, MeSH terms, keywords, PMC full-text links
• Supports MCP Prompts for systematic reviews (PICO framework, author search)
• get_paper_fulltext works for open-access PMC articles; returns helpful error for paywalled ones
• Startup takes ~5-8s via uvx (biopython + metapub dependencies)
• 3 results returned in ~4s for the CRISPR query