Fetch MCP Server

What is Fetch MCP Server?

Fetch MCP Server is a Model Context Protocol server that provides web content fetching capabilities using browser automation, OCR, and multiple extraction methods. It enables LLMs to retrieve and process content from web pages, including those requiring JavaScript rendering or employing anti-scraping techniques.

How to use Fetch MCP Server?

Install and run using Docker: build the image with docker build -t mcp-server-fetch and run with docker run --rm -i mcp-server-fetch. Configure the server in your MCP client (e.g., Roo Code or Claude App) by adding the container as an MCP server. Use the fetch tool (with required url argument and optional raw boolean) or the fetch prompt to retrieve and extract web content as markdown.

Key features of Fetch MCP Server

• Multi-method extraction including browser automation and OCR
• Automatic scoring system selects the best content result
• Handles JavaScript-rendered pages and anti-scraping techniques
• Supports PDF, DOCX, and PPTX document parsing
• Customizable user-agent via the --user-agent argument
• Automated handling of cookie consent banners and full-page screenshots

Use cases of Fetch MCP Server

• Enabling LLMs to read and analyze dynamic web pages
• Extracting content from sites that block simple scraping
• Converting complex documents (PDF, DOCX, PPTX) into markdown
• Providing reliable web content for autonomous agents and chatbots

FAQ from Fetch MCP Server

What extraction methods does the server use?

The server uses browser automation (undetected-chromedriver), OCR (pytesseract with layout detection), HTML extraction (requests/BeautifulSoup), document parsing for PDF/DOCX/PPTX, and an original markdown conversion method. A scoring system automatically selects the best result.

Do I need to install any dependencies to run the server?

The recommended installation is via Docker. The Docker image includes all necessary dependencies, so you only need Docker installed on your system.

Can I customize the user-agent used by the server?

Yes. By default, the server uses a ModelContextProtocol user-agent (autonomous or user-specified depending on the request type). You can customize it by adding --user-agent=YourUserAgent to the Docker run arguments in your configuration.

How does the scoring system work?

The scoring system awards points based on content length (up to 50), structure bonuses for well-paragraphed content (up to 20), and applies quality penalties for error messages or readability issues. Debug logging is available to track scoring decisions.

Does the server support browser automation and what does it handle?

Yes, the server uses undetected-chromedriver for browser automation. It automatically handles cookie consent banners, captures full-page screenshots, and performs OCR with layout detection.