Fetch MCP Server

1.1.2 verified Sat Apr 25 auth: no javascript

MCP server and CLI for fetching web content as HTML, Markdown, plain text, JSON, or YouTube transcripts. Current stable version 1.1.2, released regularly on npm. Key differentiators: multiple output formats, Mozilla Readability integration for article extraction, YouTube transcript support via yt-dlp or direct page extraction, and proxy support. Minimal configuration, designed for MCP ecosystem but also usable as standalone CLI. Requires Node.js >=18.

Common errors

error npx mcp-fetch-server: command not found ↓

cause Node.js version is below 18 or npm not installed.

fix

Install Node.js >=18 and ensure npm is up to date. Run: node --version && npm --version

error Error: Cannot find module 'mcp-fetch-server' ↓

cause Attempting to import the package programmatically (require('mcp-fetch-server')) instead of using CLI/MCP server.

fix

Do not try to import this package in code. Use npx mcp-fetch-server to run it as an MCP server, or use CLI commands directly.

error Failed to fetch YouTube transcript: no captions found ↓

cause Video does not have captions in the requested language, or yt-dlp is missing and direct extraction failed.

fix

Try a different language code (--lang). Install yt-dlp globally for better extraction: pip install yt-dlp

error Error: read ECONNRESET ↓

cause Network issues, proxy misconfiguration, or target server blocking the request.

fix

Check network connectivity. If using a proxy, verify proxy URL. Try a different URL. Consider adding custom headers to simulate a browser User-Agent.

Warnings

gotcha The package does not export any programmatic API; it is designed to be used via command line or as an MCP server. Attempting to require() or import it in Node.js will result in an error. ↓

fix Use npx mcp-fetch-server to run as MCP server, or use CLI commands. If you need a fetch function programmatically, use libraries like undici or node-fetch directly.

gotcha YouTube transcript extraction may fail if yt-dlp is not installed and the direct page extraction method encounters issues (e.g., non-standard player, captions not accessible). The tool falls back silently but may return empty output. ↓

fix Install yt-dlp globally (pip install yt-dlp) for more reliable transcript fetching.

gotcha fetch_readable uses Mozilla Readability which may not work well on non-article pages (e.g., homepage, video pages). It may return incomplete content or empty string. ↓

fix Use fetch_html or fetch_markdown for non-article pages; fetch_readable is designed for blog posts and articles.

gotcha Default max_length is 5000 characters, which may truncate content unexpectedly for large pages. Setting DEFAULT_LIMIT=0 can return full content but may cause memory issues for very large pages. ↓

fix Set max_length parameter appropriately per request, or configure DEFAULT_LIMIT environment variable to a higher value or 0 for no limit (with caution).

deprecated The package may change its tool names or parameters in future releases as the MCP specification evolves. Check changelog when upgrading. ↓

fix Keep up-to-date with the MCP specification and package changelog on GitHub: https://github.com/zcaceres/fetch-mcp

Install

npm install mcp-fetch-server

yarn add mcp-fetch-server

pnpm add mcp-fetch-server

Imports

mcp-fetch-server

wrong

require('mcp-fetch-server') or import from 'mcp-fetch-server' as a module

correct

npx mcp-fetch-server (CLI) or add to MCP configuration as a tool (no programmatic import)

This package is a CLI/MCP server; there is no exported JavaScript module to import. Use via npx or install globally.

Quickstart

Shows how to add the MCP server to client config and run CLI commands for markdown and YouTube transcript.

// Add to your Claude Desktop or MCP client config:
{
  "mcpServers": {
    "fetch": {
      "command": "npx",
      "args": ["mcp-fetch-server"]
    }
  }
}

// Or use as CLI:
npx mcp-fetch markdown https://example.com --max-length 10000
npx mcp-fetch youtube https://www.youtube.com/watch?v=dQw4w9WgXcQ --lang es