PubMed MCP (cyanheads)

Overview

PubMed MCP (cyanheads), packaged as @cyanheads/pubmed-mcp-server, provides an MCP interface for biomedical literature workflows. It is a third-party implementation, not PubMed itself; the supplied sources do not establish endorsement by NCBI or other upstream providers. An MCP client can connect through a local stdio process, a self-hosted Streamable HTTP server or the public endpoint documented by the project. The server exposes literature tools, a database-information resource and a research-planning prompt rather than acting as an autonomous researcher.

Inputs include PubMed query expressions, structured filters, date ranges, article identifiers, partial references and MeSH lookup terms. Outputs include search results with effective-query details, article metadata and abstracts, available full text, related-paper links and formatted references. Europe PMC search extends discovery to records such as preprints, patents and Agricola entries. Full-text retrieval tries PMC, Europe PMC and, when configured with a contact email, Unpaywall; responses identify the retrieval provider and report unavailable content or truncation.

The documented workflow supports discovery, record inspection, vocabulary refinement and reference export. Article metadata can include linked retraction and correction notices, while batch responses preserve item-level failures. These facilities help a client inspect retrieval provenance, but do not establish scientific validity or exhaustive literature coverage. Identifier conversion is limited to PMC-indexed articles, and Unpaywall text extraction is explicitly best-effort. Repository code is available under Apache-2.0; that code license does not determine article reuse rights or hosted-service terms. The supplied excerpts describe functionality and setup, not independently verified retrieval accuracy or scientific testing.

Key Features

• Search PubMed using native query syntax, structured filters and date ranges, with effective-query reporting; search Europe PMC separately for additional record types.
• Retrieve article metadata and abstracts by PMID, including MeSH terms, optional grants and linked retraction, correction or comment notices.
• Retrieve available full text from PMID, PMCID or DOI inputs through PMC, Europe PMC and configured Unpaywall fallback, with provider labels, failure reasons and truncation reporting.
• Look up MeSH headings and details, suggest spelling corrections, resolve partial citations and convert identifiers for PMC-indexed articles.
• Find similar articles, citing articles or references with named fallback providers, and export citations in APA, MLA, BibTeX, RIS or Vancouver formats.
• Expose live PubMed database metadata through `pubmed://database/info` and a parameterized four-phase `research_plan` prompt.

Use Cases

• Suggested evaluation: build an MCP-assisted literature-scoping workflow for a drug target or disease mechanism, comparing PubMed results with Europe PMC preprints.
• Suggested evaluation: inspect abstracts, MeSH annotations and linked correction or retraction notices before selecting papers for a research briefing.
• Suggested evaluation: reconcile DOI, PMID and PMCID references, then export a selected bibliography for downstream reference management.
• Suggested evaluation: retrieve accessible full text for evidence extraction while retaining provider, missing-content and truncation information.

How to Use

1. Read the project README and choose a documented connection route: public Streamable HTTP, local stdio or self-hosted HTTP. Treat these as access options for the wrapper, not PubMed service installation.
2. For hosted access, configure your MCP client to use the documented endpoint. For local deployment, follow the README’s package, runtime and client-configuration instructions rather than assuming client compatibility.
3. Review the README’s configuration section. Set NCBI contact details as appropriate; configure UNPAYWALL_EMAIL if evaluating that full-text tier, and check whether Europe PMC is enabled.
4. Start with a narrowly defined question. Use pubmed_lookup_mesh to refine vocabulary and pubmed_search_articles to retrieve records; inspect the effective query and applied filters.
5. Fetch selected metadata and available full text, checking unavailable identifiers, provider labels, linked notices and truncation. Export citations only after checking the records. As an evaluation step, compare a small sample against source records and consult the code license separately from article reuse terms.