Skip to main content
Use the platform Scrape Website tool (scrape_website) when the assistant needs text from a known public URL. It retrieves page content; it does not sign into accounts or act as a browser session.

​
Set up page reading

  1. Open the assistant’s Tool Kit → Tools & APIs and add Scrape Website.
  2. Tell the assistant which pages are relevant and what to do if the page cannot be read. For example:
    Read the supplied public product page and answer using its returned text. If the page is inaccessible or the requested detail is absent, say what you could not verify and ask for the relevant information.
  3. Save and publish the assistant.
  4. Test with a public URL beginning with https:// and ask about a distinctive sentence on that page.

​
Verify it works

Compare the answer with the actual page. The tool can return the final URL, page title, extracted text, HTTP status, and whether the content was truncated. Check the returned text for the requested detail before treating the answer as verified. Long pages may be truncated. A page that loads its content only after JavaScript runs, requires authentication, or blocks automated access may not return the text you see in your browser. Use an accessible source or your knowledge base when appropriate.

​
Input, result, and limits

The required input is url: a complete public http:// or https:// URL. The reader follows up to five redirects and reads at most 16 MiB of response data. It retrieves page content; it does not sign in, click buttons, or run a browser session. For a custom request to the reader, timeout_ms defaults to 15,000 milliseconds, with a minimum of 1,000. The total request has a 28-second budget that also includes waiting for an available reader, so a larger requested timeout does not extend that budget. Optional include_html adds the fetched source to the result; it is off unless set to true, 1, yes, or on (text values are case-insensitive). These are handler options, not additional switches in the native tool’s assignment screen.

​
Troubleshooting

  • Malformed URL: include the scheme and a complete public URL.
  • Blocked URL or private address: use a public page; internal network addresses are not supported by this tool.
  • No useful text: inspect the returned content and try the specific article URL instead of a site homepage.
  • Busy or timed out: retry the read after a short delay or use a shorter source. Do not claim the page was read when the tool failed.
Platform tools · Search the web