OfficialReference serverslocal stdio

Fetch MCP Server

Fetches a URL and converts the page to markdown for the model, in chunks.

Install

Claude Desktop

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

Copilot (VS Code)

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

Commands and configuration as given in the server’s documentation (Documentation). Replace placeholders such as API keys with your own values.

Fetch MCP Server is a Python reference server with a single fetch tool that turns web pages into markdown.

The Fetch MCP server fetches a URL and converts the page to markdown for the model, in chunks. It comes from the Model Context Protocol reference servers under the MIT license, runs locally over stdio through uvx and needs no authentication. The configuration below covers Claude Desktop and Copilot in VS Code. It respects robots.txt by default, and its README warns that it can reach local and internal IP addresses.

The tool takes max_length and start_index arguments for reading long pages in chunks and returns raw HTML on request. The user agent and proxy are configurable.

Questions

What does the Fetch MCP server do?

Its single fetch tool retrieves a web page and returns it as markdown, or as raw HTML on request.

How does the Fetch MCP server handle long pages?

The max_length and start_index arguments let the model read a long page in chunks.

Does the Fetch MCP server respect robots.txt?

Yes, by default. The user agent and proxy are configurable.

Can the Fetch MCP server reach internal addresses?

Yes. Its README warns that it can reach local and internal IP addresses.

Sources

Checked October 2, 2026.

Reference servers

View all