Search & WebOfficial

Fetch

Retrieve and convert web content to markdown for AI consumption, with support for robots.txt compliance.

Overview

The Fetch server lets your AI assistant download a public web page and read it. The page is converted to markdown (a simple text format) so the assistant gets the actual article or documentation rather than a wall of code. For example, you can paste a link to a long news story or a product manual and ask for a summary, or ask it to pull the key steps out of an online tutorial.

This is the official reference server from the Model Context Protocol project, and it is a Python program, so it runs with uvx (a tool that fetches and runs Python apps in one command) rather than npx. It only reads pages you point it at; it cannot log in to sites, fill in forms, or change anything on the web. By default it also respects robots.txt, the file sites use to say what automated tools may read.

Instructions

uvx mcp-server-fetch

Install uv onceThe server is Python-based, so you need uv (which provides the uvx command). Install it from docs.astral.sh/uv. On Windows it is one PowerShell command. After that, uvx mcp-server-fetch downloads and runs the server automatically; there is nothing else to install.
Add it to your clientIn Claude Desktop, add an entry to claude_desktop_config.json with command "uvx" and args ["mcp-server-fetch"]. In Claude Code, it is one line: claude mcp add fetch -- uvx mcp-server-fetch. No API keys or accounts are needed.
Optional flagsAdd --user-agent="YourBot/1.0" if a site blocks the default identity, or --ignore-robots-txt if you knowingly need to fetch a page that disallows bots. Most people never need either flag.
First testRestart the client and ask: "Fetch https://modelcontextprotocol.io and summarize it." You should get a clean summary. Long pages are returned in chunks. The assistant can ask for the next chunk with a start_index automatically.

Starter recipes

Real tasks you can ask an agent once this server is connected.

Summarize a long article

Paste any public link and ask for a three-bullet summary, or for the parts relevant to a decision you are making, without reading the whole page yourself.

Try

Compare two product pages

Give it the pricing pages of two services and ask for a plain-English comparison table of what each plan actually includes.

Try

Pull steps from documentation

Point it at a setup guide and ask it to extract just the steps that apply to your operating system, rewritten for a beginner.

Try

Install

uvx mcp-server-fetch

Runs on your computer through uvx, which fetches the Python package the first time.

From the project README · checked 2 Oct 2026

Owner

modelcontextprotocol

github.com/modelcontextprotocol/servers

Downloads

32.7M

PyPI · all time since Nov 2024 · as of 2 Oct 2026

Checked 2 Oct 2026

License

MIT

Official server

Access

Read-only: it can fetch any public URL you (or the assistant) give it, but cannot log in, submit forms, or modify websites. Be aware the assistant chooses URLs, so it could in principle reach internal addresses on your network. Avoid running it on machines with sensitive internal services if that worries you.

Fetch

Starter recipes

Summarize a long article

Compare two product pages

Pull steps from documentation

One AI recipe you can use.Every Monday.

One AI recipe you can use.
Every Monday.