MCP Fetch Node

🚀 Fetch MCP Server (Node.js Version)

This is the Node.js port of the official Fetch MCP Server. Please refer to the Key Differences from the Original Project section below for more details.

🚀 Quick Start

mcp-fetch-node provides an SSE interface on the default port 8080, which is available at the path /sse by default.

Node.js

npx -y mcp-fetch-node

Docker

docker run -it tgambet/mcp-fetch-node

✨ Features

• Fetch and extract relevant information from URLs.
• Respect the robots.txt (can be disabled).
• Customize the user agent.
• Convert content to Markdown format.
• Support pagination.

📦 Installation

The installation methods are as follows:

Node.js

npx -y mcp-fetch-node

Docker

docker run -it tgambet/mcp-fetch-node

💻 Usage Examples

Basic Usage

mcp-fetch-node provides an SSE interface on the default port 8080, available at the path /sse by default. You can start the server using the following commands:

Node.js

npx -y mcp-fetch-node

Docker

docker run -it tgambet/mcp-fetch-node

Advanced Usage

Custom Configuration - robots.txt

By default, the server will respect the website's robots.txt file for requests from the model (via tools), but not for user-initiated requests (via prompts). You can disable this feature by adding the --ignore-robots-txt parameter when running the command.

Custom Configuration - User Agent

By default, the server uses different user agents depending on whether the request is from the model (via tools) or user-initiated (via prompts):

# Tool invocation
ModelContextProtocol/1.0 (Autonomous; +https://github.com/tgambet/mcp-fetch-node)

# Prompt
ModelContextProtocol/1.0 (User-Specified; +https://github.com/tgambet/mcp-fetch-node)

You can customize the user agent by adding the --user-agent=YourUserAgent parameter when running the command to override the default values.

📚 Documentation

Description

A Model Context Protocol server that provides web content fetching capabilities. This server enables LLMs to fetch and process content from web pages, converting HTML to Markdown format for easier use.

The Fetch tool truncates the response, but by using the start_index parameter, you can specify the starting position for content extraction. This allows the model to read the web page in chunks until it finds the required information.

Available Tools

• fetch - Fetch a URL on the internet and extract its content in Markdown format.
  • url (string, required): The URL to fetch.
  • max_length (integer, optional): The maximum number of characters to return (default: 5000).
  • start_index (integer, optional): Start extracting content from this character index (default: 0).
  • raw (boolean, optional): Get the content without converting it to Markdown format (default: false).

Available Prompts

• fetch - Fetch a URL and extract its content in Markdown format.
  • url (string, required): The URL to fetch.

🔧 Technical Details

Key Differences from the Original Project

• This project is written in TypeScript and optimized for the Node.js runtime environment. It is suitable for situations where Python cannot be used.
• This project provides an SSE interface instead of stdio. It is more suitable for deployment as a web service, thus increasing flexibility.
• This project does not rely on the Readability.js library for content extraction. It uses a more general custom implementation suitable for handling websites other than news-related ones.

Although the API and tool descriptions are the same as the original project, you can try testing mcp-fetch-node as a drop-in replacement for the original project.

Please report any issues to the issue tracker.

📄 License

MIT

TODO

• [ ] Add user logs and progress indicators.
• [ ] Add documentation and examples.
• [ ] Perform performance benchmarking and improvements.
• [ ] Benchmark the extraction quality: Refer to https://github.com/adbar/trafilatura/blob/master/tests/comparison_small.py