mcp-defuddle

MCP server that wraps defuddle to extract clean content and metadata from web pages.

Features

• fetch_and_parse — Fetch a URL and extract its main content as Markdown or HTML. No browser required.
• parse_html — Parse a raw HTML string and extract main content with metadata.
• YouTube transcripts via defuddle's async extractors (use_async: true)
• Extracts title, author, published date, description, and cleaned content
• Optional Markdown output (default), language preference, and CSS content selector override

Installation

npx @cynosure-mcp/defuddle

Tools

fetch_and_parse

Fetches a URL over HTTP and extracts the main content using defuddle.

| Parameter | Type | Default | Description | | ------------------ | ------- | ------- | ------------------------------------------------ | | url | string | — | URL to fetch and parse | | markdown | boolean | true | Return content as Markdown | | language | string | — | Preferred language (BCP 47, e.g. en) | | use_async | boolean | true | Enable async extractors (YouTube, Twitter, etc.) | | content_selector | string | — | CSS selector to override auto-detection | | include_replies | boolean | — | Include reply/comment content |

Development

npm install
npm run build
npm start