Fetch MCP Server (Generic HTTP Requests)

Fetch Server: Arbitrary HTTP

The fetch server is a Python package distributed via PyPI and run with uvx (uv's tool-runner). It performs HTTP GETs against any URL and converts HTML responses into Markdown so the model receives clean, token-efficient text.

Config

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

Installing uvx

If you don't already have uv installed (uvx is part of the uv CLI):

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

After install, uvx mcp-server-fetch will fetch the package on first run and cache it.

What the server returns

Given a URL like https://example.com/article, the server fetches the page, runs Readability-style extraction, and returns the main text as Markdown. Compared to passing raw HTML to the model, this:

• Cuts token usage by 80–95% on most articles.
• Strips nav/footer/ad chrome that confuses summarization.
• Preserves headings and code blocks for technical content.

Limitations

• No POST/PUT/DELETE — fetch is read-only by spec.
• No JS execution — single-page apps that render content client-side return mostly empty bodies.
• No auth headers — for authenticated APIs, write a custom server (see custom stdio server) or use the dedicated server for that API.

Pairing with Brave Search

Combine fetch with Brave Search for a complete web-research workflow: Brave returns links, fetch reads each one in detail.