Fetch Official
Fetch a web page and hand the agent readable text instead of raw HTML.
- Transport
- stdio
- Package
- mcp-server-fetch
- Registry id
- —
No install snippet on purpose. A working MCP config is a command, its arguments and an environment block — the last two are where API keys live, so this catalogue never stores them and cannot publish them. Follow the link above for the authors' own instructions.
Retrieves a URL and converts it to markdown before the model sees it, which is both a token saving and a legibility one — a modern page is mostly script tags and layout wrappers.
Read this part before you wire it up
A fetch server is the classic indirect prompt injection surface: whatever it returns is attacker-controlled text arriving inside a trusted-looking tool result, and the page knows an agent is reading it. Treat everything it hands back as data to evaluate, never as instructions to follow, and be deliberate about which other tools are available in the same session — a fetch tool beside an unscoped filesystem tool is a much more interesting target than either alone.
Where it fits
Reading documentation, checking a changelog, following a link a person pasted. It is not a search engine and not a crawler.
Worth knowing
- It respects
robots.txt, which is the courtesy you want from anything fetching on your behalf. - Pages behind a login or heavy client-side rendering come back thin or empty; that is a limit of fetching rather than a bug.