Fetch a web page as clean markdown via MCP

Recipe: Fetch a URL as markdown with markfetch

Surface: markfetch v0.5.0 (npm) Transport: stdio Launch: npx -y markfetch Tool: fetch_markdown Auth: none required

How it works

1. Spawn the server via npx -y markfetch on stdio.
2. Send the MCP initialize handshake (protocol 2024-11-05).
3. Call tools/call with method fetch_markdown and {"url": "<target>"}.
4. The server fetches the page, extracts the main article content, strips nav/ads/scripts, and returns clean markdown.

Schema

{
  "name": "fetch_markdown",
  "inputSchema": {
    "type": "object",
    "properties": {
      "url": { "type": "string", "format": "uri" },
      "savePath": { "type": "string", "pattern": "^\\/", "description": "Optional absolute path to save markdown to disk" }
    },
    "required": ["url"]
  }
}

Notes

• Requires Node >= 24 (warns on 22 but runs fine).
• Follows redirects automatically.
• No cookies/auth headers sent — public pages only.
• savePath is optional; useful when output exceeds client inline caps.
• Returns isError: true with a [code] string on fetch/extraction failures.
• JavaScript-rendered SPAs return structured errors, not partial content.