biomedical-paper-triage

Biomedical Paper Triage

Turn a raw list of biomedical papers (DOIs, PMIDs, arXiv ids, or titles) into a defensible, ranked shortlist. Pairs with the consensus-tooluniverse-research evidence-grading scheme (T1-T4 x human / animal / in-vitro) so the shortlist carries both a triage rank and an evidence tier.

When to use

• You ran a search (PubMed, Consensus, a colleague's bibliography) and got back 20-200 candidate papers and need to decide which 10-20 to actually read.
• You must justify inclusion/exclusion for a mini-review or grant background and want recency, study-type, citation-signal, and access reasons recorded per paper.
• You want to avoid wasting time queuing papers that turn out to be paywalled, retracted, or non-redistributable.
• You need an access-honest list to hand to the consensus-tooluniverse-research workflow for evidence grading and download.

The MCP tools it uses

It primarily uses three tools (spec dot-name -> MCP underscore-name):

• paper.batch_resolve -> paper_batch_resolve(identifiers[]) : resolve a whole list at once into a table of canonical metadata, normalized ids (doi, pmid, arxiv, openalex), access status, OA links, and a list of unresolved items. This is the workhorse for the first pass.
• paper.access_status -> paper_access_status(identifier|doi|arxiv) : confirm or refine the honest access category for a single paper when the batch result is unknown or you need provenance.
• paper.citation_graph -> paper_citation_graph(identifier|doi, limit) : pull references and citing works from Semantic Scholar / OpenAlex to derive a citation signal and to discover high-impact neighbors worth adding to the pool.

Supporting tools when you need them: paper.resolve (single id sanity check), paper.fetch_metadata (abstract + references for closer study-type reading), paper.find_oa_pdf and paper.download_oa_pdf (only after triage, only for oa_pdf items at allowlisted hosts), and paper.summarize (structured digest from legally-available metadata).

Access categories (honest)

• oa_pdf : downloadable OA at a legit source.
• subscription : paywalled. Cite only.
• unavailable : no available full text exists.
• unknown : access could not be determined; run paper_access_status to refine.

Workflow steps

1. Gather the candidate identifiers into a single list (DOIs, PMIDs, arXiv ids, or titles). Deduplicate obvious repeats.
2. Run paper_batch_resolve(identifiers) to get one table: canonical title, year, normalized ids, access status, OA link, and the unresolved set.
3. For any item with status unknown, call paper_access_status on its best id to get an honest category and provenance. Triage unresolved items separately (often title typos or non-indexed preprints).
4. Derive a citation signal: for the top candidates by relevance/recency, call paper_citation_graph(doi, limit) and record citing-work count plus any high-impact neighbors that should join the pool.
5. Infer study type from metadata/abstract (paper_fetch_metadata when the title is ambiguous) and map it to an evidence tier using the paired consensus-tooluniverse-research T1-T4 x human / animal / in-vitro scheme (e.g. RCT / systematic review = higher human-tier; cell-line only = in-vitro tier).
6. Score and rank each paper across five axes: relevance to the question, recency, study-type strength, citation signal, and availability. Keep availability as a tag, not a hard filter, unless the user wants only readable papers.
7. Produce the ranked, access-tagged shortlist (rank, title, year, ids, study type, evidence tier, citation signal, access category, one-line reason). Hand oa_pdf items to download only after the shortlist is approved.

Concrete examples

• Resolve a known landmark: paper_resolve("10.1371/journal.pmed.0020124") returns the resolved record titled "Why Most Published Research Findings Are False".
• Confirm openly downloadable: paper_access_status(doi="10.1371/journal.pone.0000308") returns oa_pdf with a PLOS OA url, so this paper sits high on the availability axis.
• Citation neighbors: paper_citation_graph(doi="10.1371/journal.pmed.0020124", limit=25) returns references and citing works to compute citation signal and surface high-impact additions.

Limitations

• Triage scores are heuristic; a low rank is a reading-priority signal, not a quality verdict. A human reviews the final shortlist.
• Citation counts depend on Semantic Scholar / OpenAlex coverage and lag for very recent or niche work.
• Study-type inference from titles/abstracts can be wrong; verify before assigning a high evidence tier.
• unknown and unavailable items are surfaced, not silently dropped; decide their fate explicitly.

Tested invocation

paper_batch_resolve(["10.1371/journal.pmed.0020124", "10.1371/journal.pone.0000308", "10.1056/NEJMoa2034577"])
# -> table: each id resolved with canonical title, normalized ids, and access status
#    (PLOS Medicine + PLOS ONE = oa_pdf; NEJM = subscription / free_to_read_not_redistributable)

paper_access_status(doi="10.1371/journal.pone.0000308")
# -> { category: "oa_pdf", oa_url: "<PLOS OA url>", provenance: ... }

paper_citation_graph(doi="10.1371/journal.pmed.0020124", limit=25)
# -> references + citing works for the citation-signal axis