fetch-mcp is an HTTP fetching server based on the Model Context Protocol, whose main purpose is to enable MCP clients to conveniently retrieve web content and convert it into multiple formats suitable for processing.
Main capabilities include:
- fetch_html: Fetches a website and returns the raw HTML content.
- fetch_markdown: Fetches a website and converts its content into Markdown.
- fetch_txt: Fetches a website and returns plain text after removing HTML tags, scripts, and styles.
- fetch_json: Fetches a URL and returns the JSON response.
- fetch_readable: Uses Mozilla Readability to extract the main article content of a web page, returned as Markdown. This tool strips out navigation, ads, and boilerplate content, making it suitable for articles and blog posts.
- fetch_youtube_transcript: Fetches the subtitles/transcript of a YouTube video. It prefers yt-dlp (if available), otherwise extracts directly from the page. It additionally accepts a lang parameter (default "en") to select the subtitle language.
Regarding usage, the documentation states that it can be added as an MCP server to an MCP client configuration, or installed globally. It also mentions support for environment variables, with examples of customizable limits. The specific installation commands, configuration examples, and environment variable details are not fully provided in the supplied material.
Applicable scenarios include: automated workflows that need to fetch web pages through an MCP client and convert them to HTML, JSON, plain text, or Markdown; the need to use Mozilla Readability to extract article body text to remove ads and navigation; the need to obtain YouTube video subtitles; and making requests through a proxy behind a firewall.
Limitations and security features: built-in SSRF protection (blocks private/localhost addresses and DNS rebinding), provides response size limits to prevent memory exhaustion, supports pagination (max_length and start_index) as well as custom request headers. The project uses the MIT license.
