The Fetch MCP server is Anthropic’s official URL fetcher. Give Claude a URL, get back the page content as clean Markdown. It’s the simplest possible web-access primitive and one of the most useful first installs.
The server is a thin wrapper over an HTTP client and an HTML-to-Markdown converter. No browser engine, no rendering, no JavaScript. Just HTTP plus parse plus return. That simplicity is the point.
Why use it
Most “let Claude read a webpage” tasks don’t need a full browser. They need to fetch the HTML, strip the navigation, convert to Markdown, return. Fetch does exactly this and stays out of the way.
For research workflows, Fetch is the missing piece between Claude and the open web. Brave Search finds the URLs, Fetch reads them, Claude synthesises. Three tools, one prompt, full coverage.
What it actually does
A single primitive: fetch a URL and return the content. The server handles redirects, content-type detection, and HTML to Markdown conversion. It supports basic auth headers if you pass them in. Output length can be paginated for large pages.
Practical patterns:
- “Fetch this article and summarise the three main points.”
- “Read the latest release notes from this URL and tell me what changed.”
- “Pull the docs for this API and explain how authentication works.”
Gotchas
JavaScript-heavy sites return mostly empty Markdown. If the page is a single-page app that renders content via JS, Fetch sees the unhydrated HTML. For these sites you need a real browser. Firecrawl is the lighter option (managed cloud rendering); Playwright is heavier but local.
User-agent strings are basic. Some sites block default user agents. Most servers let you customise the UA; check the specific server’s config if a fetch returns 403.
The natural pairing is Brave Search plus Fetch. Brave finds the URL, Fetch reads it. Add Perplexity for cases where you want a synthesised answer rather than raw pages, and you’ve covered most web-research workflows.
For research where you’ve already got a personal library of saved sources, ContextBolt plus Fetch is even better. Claude searches your bookmarks first, fetches anything missing, then synthesises across both.
