MCP Server for Web2MD — let Claude Code, Cursor, Codex, or Claude Desktop convert webpages to clean Markdown, including the sites that block server-side crawlers.

Quick start

npm i -g web2md-mcp-server && web2md-mcp-setup

That one command detects which MCP clients you have installed and configures each one, then registers the local bridge to your Chrome. It signs itself in through the browser if you have no credentials yet — there is nothing to paste.

Then fully quit and reopen Chrome (Cmd+Q on macOS, system tray on Windows). Chrome only reads native-messaging manifests on a cold start, so closing the window is not enough. This is the most common reason setup looks like it worked and the tools are not there.

Why route through your browser

Server-side fetchers — including Firecrawl, Jina Reader, and the built-in fetch in Claude Code and Cursor — request from datacenter IPs. Reddit, X, LinkedIn and others block exactly that, and many render content client-side, so even a successful request returns an empty shell.

Your browser has neither problem: valid cookies, a real TLS fingerprint, the anti-bot challenge already passed. Measured on this setup:

Site	Result
Reddit threads and listings	complete, 4,000–6,000 chars
Hacker News front page	complete, 42,264 chars
Lobsters, Product Hunt	complete
X — trending	works (trend titles with post counts)
X — profile and search	partial: page frame, not the timeline
Stack Overflow question index	failed
GitHub trending	failed (tab timeout)

Coverage varies by site — test the ones you care about rather than assuming blanket support. A batch of 5 subreddit pages took 31 seconds (~6s per URL, processed sequentially because it is a real browser opening real tabs).

Manual configuration

The setup command above handles this for you. Use these only if you need to configure a client by hand.

Claude Desktop

Add to your Claude Desktop config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "web2md": {
      "command": "npx",
      "args": ["web2md-mcp-server"],
      "env": {
        "WEB2MD_API_KEY": "w2m_your_api_key"
      }
    }
  }
}

3. Configure Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "web2md": {
      "command": "npx",
      "args": ["web2md-mcp-server"],
      "env": {
        "WEB2MD_API_KEY": "w2m_your_api_key"
      }
    }
  }
}

Windows Notes

npx not working?

If npx web2md-mcp-server fails on Windows with "not recognized", use node directly:

Install globally: npm install -g web2md-mcp-server
Find the install path: npm root -g
Use this config instead:

{
  "mcpServers": {
    "web2md": {
      "command": "node",
      "args": ["C:\\Users\\<user>\\AppData\\Roaming\\npm\\node_modules\\web2md-mcp-server\\dist\\index.js"],
      "env": {
        "WEB2MD_API_KEY": "w2m_your_api_key"
      }
    }
  }
}

Replace <user> with your Windows username, or run npm root -g to find the exact path.

Claude Desktop: use the standard installer

The Microsoft Store version of Claude Desktop cannot launch local MCP servers due to UWP sandboxing. Install Claude Desktop from claude.ai/download instead.

MCP Tools

`convert_url`

Convert a single webpage URL to Markdown.

Input: { url: "https://example.com/article" }
Output: { markdown: "# Article Title\n...", metadata: { title, wordCount, readingTime } }

`batch_convert`

Convert multiple URLs at once (up to 50).

Input: { urls: ["https://...", "https://..."] }
Output: [{ url, markdown, metadata }, ...]

`agent_convert` / `agent_batch_convert`

Convert URLs through the Chrome extension's real browser session (requires Agent Bridge setup). Works on Reddit, login-protected sites, and JS-rendered pages.

Agent Bridge

For full browser-based conversion (bypasses anti-bot), see Agent Bridge docs.

Keywords

FAQs

What is web2md-mcp-server?

Is web2md-mcp-server well maintained?

Package last updated on 06 Sep 2026

web2md-mcp-server