PubMed Articles Scraper — Search Results with Full Citation Metadata

Search PubMed and get structured article records: title, full author list, journal and ISSNs, publication dates, volume/issue/pages, DOI and PMC ID, publication types and citation counts.

Keyless NCBI E-utilities API. No login, no browser, no proxy needed.

Input

Field	Type	Default	Description
searchQueries	array	required	Full PubMed syntax, e.g. crispr AND cancer[Title]
publishedFrom / publishedTo	string	—	YYYY/MM/DD
dateType	string	pdat	Which date the filter matches — see below
sort	string	relevance	pub_date, Author, JournalName
maxItemsPerQuery	integer	200	Capped by the API at 9,999
apiKey	string (secret)	—	Free from NCBI; raises 3 req/s to 10
contactEmail	string	—	NCBI asks callers to identify themselves

Output

{
  "_input": "crispr AND cancer",
  "_source": "S1-eutils-esummary",
  "_scrapedAt": "2026-09-09T13:44:12Z",

  "pmid": "42711491",
  "url": "https://pubmed.ncbi.nlm.nih.gov/42711491/",
  "title": "Chemogenomic maps reveal a PRDX1-dependent iron-damage axis…",

  "authors": ["O'Loughlin TA", "Arab A", …],
  "authorCount": 14,
  "firstAuthor": "O'Loughlin TA", "lastAuthor": "Bhatt DM",

  "journal": "Nat Chem Biol",
  "journalFull": "Nature chemical biology",
  "issn": "1552-4450", "eIssn": "1552-4469", "nlmUniqueId": "101231976",

  "pubDate": "2026 Sep 8",          // ← what a date filter matches
  "ePubDate": "2026 Sep 8",
  "sortPubDate": "2026/09/08 00:00", // ← PubMed's sort key; see below
  "docDate": null,
  "volume": "", "issue": "", "pages": "",
  "language": ["eng"],
  "publicationTypes": ["Journal Article"],

  "doi": "10.1038/s41589-026-02312-z",
  "doiUrl": "https://doi.org/10.1038/s41589-026-02312-z",
  "pmcId": null,
  "elocationId": "doi: 10.1038/s41589-026-02312-z",
  "articleIds": { "pubmed": "42711491", "doi": "…" },

  "pmcRefCount": 12,
  "recordStatus": "PubMed - as supplied by publisher",
  "publicationStatus": "aheadofprint"
}

Three things worth knowing

pubDate and sortPubDate are different dates, and the filter matches the first. PubMed's sort key reflects later revisions: a 2010 article updated in 2024 sorts as 2024. Filtering a run to 2010 returned rows whose pubDate was 2010 for 37 of 40 — while their sortPubDate was mostly 2013 and 2014. Compare against pubDate, or the filter will look broken when it is working exactly as specified. dateType lets you filter on the Entrez (edat) or modification (mdat) date instead.

PubMed serves at most 9,999 records per query. retstart cannot exceed 9998, and retmax is silently clamped: ask for 10,000 and you get 9,999, no warning. Narrowing by date range is the way past it — the actor logs the true match count and says so when you have asked for more than can be served.

The error path is not valid JSON. Overrunning the ceiling returns HTTP 200 with an ERROR key, and the message embeds a raw newline inside a JSON string. Strict json.loads refuses it outright with Invalid control character, so the actual explanation is lost and the failure looks like a transport error. The success path parses fine — meaning a parser can pass every test and still fail on the first real failure. This actor parses leniently and reports the message.

Rate limits and courtesy

NCBI documents 3 requests a second without a key and 10 with one, and asks callers to identify themselves so they can get in touch about unusual usage rather than simply blocking it. The actor paces itself to the documented rate, sends a tool identifier, and passes your contactEmail when you give one. Neither is authentication; both are the terms of use.

Errors

_error	Meaning
invalid_input	Empty query
no_results	The search ran and matched nothing
api_error	A 200 carrying an ERROR field — usually the 9,999 ceiling
unexpected_shape	A 200 without esearchresult or result
blocked	Every TLS profile was refused
network_error	The ladder never reached the server

If every query fails, the run itself fails rather than reporting success over an empty dataset.