Yes. It is an open-source reference server in the official Model Context Protocol repository, licensed MIT, with no paid tier.

Does it support remote/OAuth?

No. It runs locally over stdio via uvx and requires no authentication; it simply makes outbound HTTP requests on the agent's behalf.

How is it different from a full browser server like Playwright?

Fetch does a single HTTP GET and converts the response to markdown. It does not run JavaScript, click, or navigate, so it is far lighter than a browser-automation server but cannot render client-side-rendered pages.

Fetch MCP server

OfficialAnthropic (reference)86,581Config last verified Jun 1, 2026

Reference MCP server that fetches a URL and converts the page to clean, token-efficient markdown for the agent to read.

Fetch is one of the active reference servers maintained in the official Model Context Protocol repository. It gives an agent a single, focused capability: retrieve the content of a web page and convert the HTML into clean markdown so the model can read it without wading through scripts, navigation chrome, and styling noise. Because the output is markdown rather than raw HTML, it is dramatically more token-efficient, which matters when an agent is reading documentation, articles, or API references mid-task.

The fetch tool supports chunked reading: max_length caps how much content comes back in one call and start_index lets the agent page through a long document across several calls instead of blowing its context window on a single huge page. A raw flag returns the unprocessed HTML when the markdown conversion would lose something important. The server runs locally over stdio, launched with uvx, and exposes flags to ignore robots.txt, set a custom user agent, or route through a proxy. It is the canonical way to let a coding agent pull in a web page it needs without giving it a full browser.

Quick install

Copy-paste configs are provided for all 8 supported clients. Pick your client below.

Add to ~/.claude.json

~/.claude.json

json

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

Or via CLI

bash

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

Step-by-step guides

Available tools

Tool	Description
fetch	Fetches a URL from the internet and extracts its contents as markdown, with optional max_length, start_index for paging through long pages, and a raw flag to return unprocessed HTML.

What you can do with it

Read documentation mid-task

When an agent hits an unfamiliar API or library, it fetches the relevant docs page as markdown and continues coding without you pasting anything in.

Page through a long article

For a long page the agent reads the first chunk, then calls fetch again with a higher start_index to continue, keeping each response small enough to fit the context window.

FAQ

Is it free?: Yes. It is an open-source reference server in the official Model Context Protocol repository, licensed MIT, with no paid tier.
Does it support remote/OAuth?: No. It runs locally over stdio via uvx and requires no authentication; it simply makes outbound HTTP requests on the agent's behalf.
How is it different from a full browser server like Playwright?: Fetch does a single HTTP GET and converts the response to markdown. It does not run JavaScript, click, or navigate, so it is far lighter than a browser-automation server but cannot render client-side-rendered pages.

← Browse all dev-tools servers