MCP servers / Fetch

Fetch MCP server: setup for Claude Code, Claude Desktop and Cursor

The Fetch server has one tool: it downloads a web page and returns it as readable text, converted to markdown. It is the simplest way to let your AI read a URL you give it.

Unlike Playwright it does not run a browser or click anything. It fetches the page and hands back the text.

Runs with

uvx (uv, for Python)

Package

mcp-server-fetch on PyPI

Needs

Python tooling: it runs through uvx, which comes with uv.

Tools

1 (not declared as read-only or not)

Source

modelcontextprotocol/servers

Last checked

2026-09-20 (server reported version 1.30.0)

Config for Fetch

Generated by the same code as the free builder, so it matches what you get there. Pick your client, then copy.

Claude Code

Save as .mcp.json in your project root. Claude Code asks you to approve it the first time.

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

Or add it from a terminal instead (single quotes work in bash, zsh and PowerShell; more on this command):

claude mcp add fetch -- uvx mcp-server-fetch

Claude Desktop

Merge into claude_desktop_config.json (Settings → Developer → Edit Config), then fully quit and restart Claude Desktop.

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

Cursor

Save as .cursor/mcp.json in your project, or ~/.cursor/mcp.json for all projects.

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

What it can do

These is the tool the server reported when we started it and asked on 2026-09-20. The labels come from the hints each tool declares about itself.

• fetch
  Fetches a URL from the internet and optionally extracts its contents as markdown.

Watch out for

• The maintainers warn that it can reach local and internal network addresses. On a company network, that can include internal pages, so be careful what it is pointed at.
• By default it obeys a site's robots.txt when the AI is the one asking. The README documents --ignore-robots-txt to turn that off, and --user-agent and --proxy-url to change how it identifies itself and route through a proxy. Leave robots.txt on unless you have a reason.
• The server does not declare itself read-only or not in its tool definition, but a single page fetch is a request from your machine, so it can still hit any address you can reach.

Try asking

• “Fetch https://example.com and summarize the page in three bullet points.”
• “Read the changelog at this URL and tell me what changed in the last two releases.”
• “Fetch this API reference page and list its endpoints.”

Related servers

• Playwright: Drive a real browser to test pages and take screenshots.
• Firecrawl: Scrape and crawl websites into clean text.