Chunking is the point: the response is capped and resumable, so a long document can be walked through until the part that answers the question appears, rather than filling the context in one go.
A web fetcher that converts HTML to markdown before returning it. Responses are truncated by design, and a start index lets the caller move through a long page piece by piece rather than asking for all of it.
- A URL fetched and returned as markdown, with the page furniture stripped out
- A maximum length per response, so a long page does not arrive as one block
- A start index for continuing from where the last chunk ended
- Raw content on request, when the markdown conversion is not what you want
- A prompt as well as a tool, for the plain case of fetching one URL and reading it
Uv, which runs the published package without an install step, or pip, or the published container image. Node.js is optional and worth having: with it present, a more robust HTML simplifier is used. The server can reach local and internal addresses as well as public ones, so run it where that is acceptable.
One command — pip install mcp-server-fetch