Find research papers and verify DOI metadata with an AI agent

Run the workflow

1. 1. Define the question and the scope

   Write the question, relevant terms and the time period you want to investigate. Decide whether you need a historical source, recent work or both. For example, a brief on information theory can begin with foundational terminology and a known historical paper, rather than assuming the most recent result is the best source.

2. 2. Use an overview to find vocabulary

   On Wikipedia Research, use search_articles to identify a specific English article and read_article with its title. Keep the returned source and revision links. Check the disambiguation flag before proceeding. Use the overview to collect names and search terms; sampled external links and extracted text are not an independently verified bibliography.

3. 3. Search scholarly metadata with an explicit cutoff

   On Crossref Research, call search_publications with your topic or author, a small limit, and optional fromYear and toYear publication filters. Review titles for relevance and preserve the DOI links. A search returns up to ten records, not an exhaustive literature review. Change the query when results miss your question; do not treat the first shortlist as complete coverage.

4. 4. Verify each selected identifier

   Call publication_details with the plain DOI, without a DOI URL or whitespace. Compare the returned title, author names, journal and publication date with the reference you intend to use. Preserve the missingMetadata list. A record match supports bibliographic identification; it does not show that the paper supports a particular sentence.

5. 5. Read the publication before writing the claim

   Open the DOI source link, locate a version you are entitled to read, and check the relevant passage, methods and limitations. Keep the publication version and location supporting each factual claim in your brief. Do not ask the agent to reconstruct an abstract or findings from a title. Leave a claim unresolved when the underlying work has not been read.

What a useful result looks like

A useful source ledger separates three things: bibliographic fields observed in a record; your hypothesis about relevance; and claims checked against a publication you have actually read. Record a source URL for each entry, unresolved fields and the next reading task. Citation counts can help describe the deposited record but are not a quality ranking.

Know the limits

Wikipedia Research covers English article extracts, which may omit tables, formulas and citations. Crossref Research provides bounded bibliographic metadata for Crossref-registered works, not abstracts, full text or a complete literature search. Publisher-deposited fields can be missing, incorrect or outdated. Neither connector independently checks the truth of a publication, resolves every citation or writes a guaranteed-correct bibliography. Both require ToolCargo authentication and use the shared call quota; no Wikipedia or Crossref provider key is required.