| description | Download a PDF from a URL and extract its full text — ideal for research papers, reports, and documentation. Returns page count and extracted text. |
| name | pdf_extraction |
| source | built-in |
| tags | ["pdf","research","papers","documents","extraction"] |
| tools | ["url_fetch"] |
When to use
Apply when the work item requires reading the actual content of a research paper,
technical report, regulation document, methodology specification, or any content
published as a PDF. Use when web_browse returns only a landing page or abstract
and the full text is needed. Signals: "read the paper", "extract from the report",
"check the methodology", "find the exact wording in".
How to execute
- Confirm the URL points directly to a PDF (ends in .pdf or is a direct download
link from a publisher/repository), not to an HTML landing page.
- Call fetch_url(url) — the tool will download and extract the text automatically.
- Read the returned text in full before extracting anything.
- Identify the sections relevant to the task: abstract, methodology, results,
appendix. Skip unrelated sections to stay within budget.
- Extract verbatim quotes for any key definitions, thresholds, equations, or findings.
- Note the document title, authors, year, and publisher for citation purposes.
Output contract
Produce a citation block (title, authors, year, URL) and a structured extraction
containing: relevant section names, verbatim quotes for all critical facts, and a
plain-language summary of what the document contributes to the task.
Quality bar
- Never paraphrase quantitative thresholds, legal definitions, or formula
components — always quote verbatim.
- Record the full citation (not just the URL) for every document extracted.
- If extraction fails or returns garbled text, report this explicitly.
- Only extract sections relevant to the task — do not dump entire documents.
Pairs well with
- web_search
- web_browse
- dataset_inspection
