Fetch

Fetch is the simplest possible bridge between an agent and the open web: given a URL, it retrieves the page and converts its HTML into clean Markdown, trimmed of navigation chrome and script noise so the content is cheap to read in context. It accepts optional max_length and start_index arguments for paginating through long pages, and a raw flag when you need the unprocessed response instead of the Markdown conversion.

It’s the right tool when an agent just needs to read a page — pull in documentation, check a changelog, or ground a research answer in a live source — rather than click through a multi-step flow. For that heavier interaction (logins, forms, JavaScript-rendered content), pair it with a full browser-automation server like Playwright instead.

One thing to flag for anyone deploying this in a shared or sensitive environment: the server can reach local and internal IP addresses by default, so treat it the same way you’d treat any outbound-HTTP-capable tool and constrain network access at the infrastructure level if that matters for your deployment.