PubMed¶

PubMed indexes biomedical and life-science literature via NCBI. Disabled by default — enable it for biomedical queries.

Implementation: src/retrieval/providers/pubmed.py

HTTP API¶

Attribute	Value
Search URL	`GET https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi`
Fetch URL	`GET https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi`
Query params	`db=pubmed`, `term`, `retmax`, `sort=relevance`
Authentication	`NCBI_API_KEY` (optional; raises rate limits)
Timeout	60s (search), 15s (health)
Retries	3 with exponential backoff
Rate limiting	HTTP 429 → sleep `Retry-After`

Search is a two-step flow: esearch returns relevance-sorted PMIDs, then a single efetch call hydrates all of them as XML — including abstracts, which NCBI's JSON esummary endpoint does not provide.

Normalization¶

PubmedArticle XML records map to RetrievedPaper:

PubMed field	`RetrievedPaper` field
`ArticleTitle`	`title`
`Abstract/AbstractText[]`	`abstract` (sections joined)
`Journal/JournalIssue/PubDate/Year`	`year`
`Journal/Title`	`venue`
PMID article page	`url`
`ELocationID[@EIdType="doi"]`	`doi`
`AuthorList/Author`	`authors` (fore + last name)
`KeywordList/Keyword`	`keywords`

Configuration¶

# config/providers.yaml
providers:
  pubmed:
    enabled: true

RA_RETRIEVAL__PROVIDERS__PUBMED__ENABLED=true
NCBI_API_KEY=...   # optional