| description | Navigate to and extract content from specific web URLs. Good for reading data portals, documentation, and landing pages. |
| name | web_browse |
| source | built-in |
| tags | ["browse","web","scraping"] |
| tools | ["web_browse"] |
When to use
Apply when you have a specific URL that you need to read in full — a data portal,
regulation document, methodology page, research paper landing page, or any page
returned by a web_search that warrants deeper reading. Use after web_search has
identified promising sources. Do not use to discover URLs (use web_search for that).
How to execute
- Confirm the URL is the primary/authoritative source, not a secondary summary page.
- Browse the URL and read the returned text in full before extracting anything.
- Identify the key sections relevant to your task — ignore navigation, footers, ads.
- Extract verbatim quotes for any critical facts, figures, or definitions.
- If the page links to downloadable documents (PDFs, datasets), note the direct URLs
for follow-up with pdf_extraction or dataset_inspection.
- Note the page title, last-updated date if visible, and the organisation responsible.
Output contract
Produce a structured note in the proof field with: source URL, page title, publishing
organisation, date retrieved, extracted key facts with verbatim quotes, and any
follow-up URLs identified for further inspection.
Quality bar
- Extract verbatim text for any figures, thresholds, or legal/regulatory language —
do not paraphrase critical quantitative claims.
- Record the exact URL browsed, not just the domain.
- If the page fails to load or returns an error, report this explicitly.
- Do not infer meaning not present in the page text.
Pairs well with
- web_search
- pdf_extraction
- dataset_inspection
