https://example.com/docs includes https://example.com/docs/setup. It does not include https://example.com/docs-other, a different subdomain, or pages requiring login. URLs cannot contain query strings or fragments. IP addresses, custom ports, and private sites are not accepted.
How answers are generated
For each question, Claude searches for relevant pages, includes registered URLs as candidates, and retrieves page content. The final answer uses only in-scope content and citations successfully retrieved for that question. Search-result descriptions alone are not used to create answers. If no citable page content is available, the bot does not return a regular answer. To ensure a deeper page that is not indexed by search is considered, register that page’s URL too. Pages requiring JavaScript execution or blocking retrieval may not be usable. The service does not train on your site in advance, accept file uploads, or ingest content into a vector database.You can currently save up to 10 source URLs. The settings screen may show a maximum of 20, but 11 or more URLs cannot be saved.