defuddle-fetch

The defuddle-fetch MCP server provides web content extraction by leveraging the Defuddle library. It replaces the default fetch tool with a more capable processor that cleans up page clutter, preserves meaningful metadata (such as title, author, and publication date), and converts content into readable markdown (with an option to output raw HTML when needed). The fetch tool supports pagination through start_index and max_length, and can return either markdown-formatted content or raw content depending on the request parameters. This makes it suitable for feeding AI models with clean, structured web content for summarization, analysis, or downstream tasks.

To use the server, send a fetch request to a URL via the fetch tool. You can specify max_length to limit the number of characters returned and start_index to skip to a particular position in the content. If you prefer the pure HTML output (for debugging or specialized processing), set raw to true. The response includes extracted metadata such as the page title, author, publication date, word count, domain, and processing time, enabling richer context for models. The server is designed as a drop-in replacement for the default fetch MCP server, offering better extraction quality on modern pages thanks to Defuddle's extraction capabilities and site-specific extractors.

Prerequisites:

• Node.js and npm installed on your machine
• Optional: Claude Desktop (if you plan to configure via Claude Desktop), or run the MCP server standalone

Installation steps:

1. Install Node.js and npm (if not already installed)

   • Visit https://nodejs.org/ and follow the installation instructions for your OS

2. Install the MCP server (via npm/publish or development clone)

   • For production usage via npx (as shown in config):
     • You can rely on the npm-published package: defuddle-fetch-mcp-server
   • For development/local testing:
     • git clone https://github.com/domdomegg/defuddle-fetch-mcp-server.git
     • cd defuddle-fetch-mcp-server
     • npm install
     • npm run build

3. Run or configure the server

   • If using npx (preferred for quick start): no local build is required; the CLI will fetch the package when starting via your MCP config
   • If running locally after build, you can start the server (example):
     • node dist/index.js

4. Configure Claude Desktop (example)

   • In claude_desktop_config.json, add the server under mcpServers: { "mcpServers": { "defuddle-fetch": { "command": "node", "args": ["/path/to/clone/defuddle-fetch-mcp-server/dist/index.js"] } } }

Notes:

• The server is designed to be a drop-in replacement for the default fetch MCP server, so you can swap it in with the existing configuration without major changes.
• If you plan to publish or reuse the package, ensure you have the correct version of Defuddle and any required dependencies installed in your environment.

Tips and common issues:

• Ensure Node.js and npm versions are compatible with the defuddle-fetch-mcp-server package.
• When using Claude Desktop, verify the path to dist/index.js after building the project.
• The fetch tool returns metadata such as title, author, publication date, word count, and domain; use these fields to enrich model prompts and provenance.
• If you encounter network or parsing issues on certain sites, try the raw output (raw: true) to bypass markdown conversion for debugging.
• The max_length parameter defaults to a sensible limit; increase only if your model needs longer content and be mindful of token limits.