Defuddle

MCP server that extracts the main content from web pages using defuddle. Returns cleaned Markdown or HTML with metadata (title, author, description, etc.), stripping away ads, navigation, sidebars, and other clutter.

No authentication required — defuddle runs locally as a CLI tool.

Tools

Tool	Description
defuddle_url	Fetch a URL and extract its main content
defuddle_html	Extract main content from raw HTML

Both tools return a structured result with: title, author, description, domain, content, word_count, and other metadata fields. Set markdown=false to get HTML instead of Markdown.

Setup

Requires Python >= 3.10 and Node.js >= 18.

uv sync

The defuddle npm package is installed automatically on first server start.

Run

uv run python -m src.main

The server starts on http://127.0.0.1:8080/mcp.

Test

uv run python -m src.client