Fetch MCP Server

The Fetch MCP Server is part of the official MCP server collection maintained by Anthropic. It gives your AI assistant the ability to read web pages by converting HTML to clean markdown — making web content digestible for LLMs that otherwise can't access the internet.

Why you need this

Without the Fetch server, your AI can only work with content you paste into the chat. With it, the AI can read documentation, check live APIs, scrape data, and research topics autonomously. It's one of the first servers most developers install after Filesystem.

How content extraction works

The server fetches the target URL, converts the HTML to markdown (stripping navigation, ads, and boilerplate), and returns paginated results. The start_index parameter lets the AI read long pages in chunks:

jsonCOPY
{
  "name": "fetch",
  "arguments": {
    "url": "https://docs.example.com/guide",
    "max_length": 5000,
    "start_index": 0
  }
}

To fetch subsequent sections, the AI increases start_index by max_length and calls again — effectively paginating through the document.

Python requirement

This server is written in Python, not Node.js. You need either uvx (recommended — no install needed) or pip install mcp-server-fetch. If you don't have Python/uv installed, the server won't start.

Quick install with pip:

bashCOPY
pip install mcp-server-fetch

Then use python -m mcp_server_fetch as the command instead of uvx.

Docker installation

jsonCOPY
{
  "mcpServers": {
    "fetch": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "mcp/fetch"]
    }
  }
}

Security warning

This server can access local and internal IP addresses. If you're running it in a shared or production environment, be aware it could be used to reach internal services. For most local development setups, this is not a concern.

Customization

Ignore robots.txt: Add --ignore-robots-txt to the args array if you need to fetch pages that block automated access.

Custom user agent: Add --user-agent=YourAgent to the args for sites that block the default MCP user agent.

Corporate proxy: Add --proxy-url=http://your-proxy:8080 to route requests through a corporate proxy.

Common issues

Windows timeout errors Add PYTHONIOENCODING to your env to fix character encoding issues on Windows:

jsonCOPY
{
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": ["mcp-server-fetch"],
      "env": {
        "PYTHONIOENCODING": "utf-8"
      }
    }
  }
}

"uvx not found" Install uv first: curl -LsSf https://astral.sh/uv/install.sh | sh (macOS/Linux) or powershell -c "irm https://astral.sh/uv/install.ps1 | iex" (Windows).

Fetched content is empty or garbled Some sites serve content via JavaScript rendering which the Fetch server can't execute. It only processes static HTML. For JS-heavy sites, consider a browser-based MCP server instead.