Fetch MCP Server

What it does

The fetch server retrieves a URL and extracts readable content as markdown for a model. A max_length limit truncates output, and start_index lets the model continue reading a long page in chunks. A raw option returns content without markdown conversion.

By default the server follows robots.txt for model initiated tool calls, but not for user initiated prompt calls. Options can disable robots handling, set a custom user agent, or route requests through a proxy.

The README carries an explicit caution: the server can reach local and internal IP addresses, which can expose sensitive data if it is used without care.

Configuration

Run with uvx mcp-server-fetch, install with pip install mcp-server-fetch, or use the mcp/fetch Docker image.

{
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": ["mcp-server-fetch"]
    }
  }
}

Source: https://github.com/modelcontextprotocol/servers/tree/main/src/fetch. Placeholders only. Replace them in your own environment, and never commit real keys to a repository.

Verified tools and features

• fetch tool Config checked

  Fetches a URL and returns extracted content as markdown, with max_length, start_index and raw options.

• fetch prompt Config checked

  Prompt form of the fetch capability that takes a required url argument.

• robots.txt handling Config checked

  Obeys robots.txt for model initiated requests by default, with an --ignore-robots-txt option to disable it.

• Proxy and user agent options Config checked

  Supports --proxy-url and --user-agent arguments for routing and identification.

Security notes