Fetch and Convert Web Content to Markdown

defuddle-fetch-mcp-server is a Model Context Protocol server that fetches a URL and converts it to clean, readable markdown using the Defuddle library, extracting the page's main content along with title, author, publication date, word count, and domain metadata. It is a drop-in replacement for the official MCP fetch server, which uses Mozilla's Readability instead of Defuddle.

Use it in place of the default fetch MCP server when web pages come back cluttered or poorly extracted - the project's own comparison shows Readability struggling on a GitHub issue page (pulling in raw avatar markup and navigation noise) where Defuddle extracts clean, readable content instead. It is specifically built around Defuddle's model of custom per-site extractors, so it should do particularly well on sites Defuddle has tuned for, GitHub among them. It is not a general web-browsing or interactive-page tool - it fetches and extracts a single URL's content, with optional pagination for long pages, rather than driving a full browser session.

• Better content extraction: strips webpage clutter while preserving the page title and key metadata.
• Flexible output: returns markdown by default, or raw HTML when the raw parameter is set.
• Chunked reading: paginate long content with start_index and max_length parameters.
• Rich metadata: title, author, publication date, word count, domain, and processing time are returned alongside the extracted content.

The single fetch tool takes url (required), plus optional max_length (default 5000 characters), start_index (default 0), and raw (default false).

Generate the right client config (Claude Code, Claude Desktop, Cursor, Cline, VS Code, and more) via the project's install-mcp link, or add it to Claude Desktop manually after building from source:

{
  "mcpServers": {
    "defuddle-fetch": {
      "command": "node",
      "args": [
        "/path/to/clone/defuddle-fetch-mcp-server/dist/index.js"
      ]
    }
  }
}

Building from source requires Git and Node.js: clone the repository, run npm install, npm run test, then npm run build before pointing a client at dist/index.js. Pull requests are welcomed, and releases follow semantic versioning: a maintainer bumps the version with npm version, pushes tags, and GitHub Actions publishes the result to the NPM registry.

Anyone using an MCP client that needs to fetch and read web pages - especially GitHub pages and other sites where the default Readability-based fetch server produces noisy or broken markdown - and wants a straightforward swap-in replacement rather than a new workflow to learn.