Summary
A page reader tuned for context budget rather than fidelity.
Every design choice here aims at returning fewer tokens for the same article: main-content detection, attribute stripping, optional links and images. It fetches publicly accessible pages, so anything behind a login or an anti-bot wall is out of scope.
What it is
A server that fetches a public URL, extracts the main content and converts it to markdown, deliberately keeping the result small.
What you get
- Main-content extraction using semantic HTML detection, with navigation, ads and sidebars removed
- HTML attributes stripped and excess whitespace collapsed, so less of the page becomes context
- Links and images included or excluded per call
- Redirects followed, with the final URL reported back
- One-second intervals between requests, a proper user-agent header and timeout handling
Requirements
Python 3.12 or higher and uv. Runs straight from the repository with uvx, so there is nothing to clone first.
Setup effort
One command — uvx git+https://github.com/bhubbb/mcp-fetch-as-markdown