Fetch MCP Server

Overview

Fetch retrieves a URL and converts its HTML to markdown so a model can read web content without also having to parse raw HTML. It's one of the official Model Context Protocol reference servers.

How it works

1. The single fetch tool takes a url, an optional max_length (default 5000 characters), a start_index to resume reading further into a page, and a raw flag to skip markdown conversion.
2. Because responses are truncated, start_index lets a model page through a long document across several calls until it finds what it needs.
3. The server respects robots.txt for model-initiated tool calls (not for user-initiated prompts), and both the user agent and this behavior are configurable via --user-agent and --ignore-robots-txt.
4. Its own README flags a real caution: this server can reach local/internal IP addresses, which is a security consideration when giving an agent open fetch access.

Examples

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

Installation

You need uv (recommended) or Python with pip.

1. Add the server to your MCP client's config using the block in the sidebar (uvx mcp-server-fetch).
2. Restart your client so it picks up the new server.
3. Optional: pass --user-agent or --ignore-robots-txt if you need to change its default fetching behavior.
4. This server can reach local/internal IP addresses, only give it to agents and prompts you trust.