Read URL

Fetches a page and strips the HTML down to readable text. Search Engine finds a link; read_url opens it — so the agent can answer from the page itself instead of a one-line snippet.

Long pages come back one slice at a time: the result ends with a [truncated at N of M chars …] note, and calling again with that start_char returns the next slice. include_links adds the page's links as absolute URLs when the next step is to follow one.

Outbound URLs go through the same allowed_domains gate as Curl Request. A search result is normally not on the allowlist, so the tool answers with the blocked host and the exact command to permit it — a trusted sender replies /allow <domain> and the domain stays allowed from then on. The agent is told not to retry until that happens.

Unlike curl_request it verifies TLS, follows at most 5 redirects, caps the download at 5 MB, and refuses binary responses (PDF, images) rather than dumping bytes. For API calls that need custom methods, headers, or a request body, use Curl Request instead.