Firecrawl-Compatible Stealth Web Scraper MCP Server

A lightweight, drop-in replacement for the core Firecrawl MCP tools (firecrawl_scrape, firecrawl_crawl, and firecrawl_map). Built for AI agents (Cursor, Claude Desktop, autonomous swarms) that need clean, LLM-ready markdown without paying for blocked requests or running into aggressive edge bot filters.

---

What This Solves

1. Strict Pay-Per-Success Billing: Standard scraping APIs consume your credits even when a site returns HTTP 403 or times out. This server only charges when valid DOM content is extracted. If a request is blocked or unreachable, you pay $0.00.
2. Resilient TLS Fingerprinting: Direct HTTP clients (curl, python-requests, default node fetch) get blocked by modern edge firewalls before reaching the page. This engine emulates realistic browser TLS Client Hello signatures (JA3/JA4) and HTTP/2 frame ordering.
3. Passive URL Mapping (Zero Target Probing): firecrawl_map queries passive threat intelligence indexes (AlienVault OTX) to discover indexed URLs and endpoints under a target domain without sending active HTTP probes to the target server.
4. Dual Settlement Rails: Connect either via Apify Pay-Per-Event ($0.005/scrape) or natively via Coinbase x402 micropayments ($0.01 USDC on Base).

---

Supported Tools

1. firecrawl_scrape

Scrapes a single webpage and returns clean, LLM-ready Markdown.

Parameters:

• url (string, required): The target webpage URL.
• formats (array, optional): ["markdown", "html"] (default: ["markdown"]).
• onlyMainContent (boolean, optional): Strips navigation bars, footers, scripts, and cookie banners (default: true).
• waitFor (integer, optional): Milliseconds to wait for client-side JavaScript rendering (default: 0).

Response Shape:

{
  "success": true,
  "data": {
    "markdown": "# Page Title\n\nArticle content...",
    "metadata": {
      "title": "Page Title",
      "description": "Page meta description",
      "language": "en",
      "sourceURL": "https://example.com/article",
      "statusCode": 200,
      "wordCount": 420
    }
  }
}

2. firecrawl_map

Discovers URLs belonging to a domain using passive intelligence without active probing.

Parameters:

• url (string, required): Domain or URL to map (e.g. example.com).
• limit (integer, optional): Maximum URLs to return (default: 100).

Response Shape:

{
  "success": true,
  "links": [
    "https://example.com/pricing",
    "https://example.com/docs",
    "https://example.com/blog/post-1"
  ]
}

3. firecrawl_crawl

Crawls linked pages on a website up to a designated depth.

Parameters:

• url (string, required): Starting URL.
• maxDepth (integer, optional): Maximum link depth (default: 2).
• limit (integer, optional): Maximum pages to scrape (default: 10).

---

Installation & Setup

Claude Desktop

Option A: Native UI (Recommended)

1. Open Claude Desktop → Settings → Connectors
2. Click Add custom connector
3. Paste: https://neon-innovation-lab--firecrawl-stealth-mcp.apify.actor/mcp

Option B: Config File (claude_desktop_config.json)

Claude Desktop's config file runs local commands, so remote HTTPS servers connect via the official mcp-remote stdio bridge:

{
  "mcpServers": {
    "firecrawl-stealth": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://neon-innovation-lab--firecrawl-stealth-mcp.apify.actor/mcp"]
    }
  }
}

Cursor IDE

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "firecrawl-stealth": {
      "type": "streamable-http",
      "url": "https://neon-innovation-lab--firecrawl-stealth-mcp.apify.actor/mcp"
    }
  }
}

Autonomous Agent Integration (x402 Micropayments)

Autonomous agents calling this MCP server can pay dynamically using the x402 protocol:

1. Make unauthenticated tool call to /mcp.
2. Receive HTTP 402 challenge with payment parameters (0.01 USDC on Base eip155:8453).
3. Retry request with signed payload in _meta["x402/payment"].
4. Settlement occurs only if extraction succeeds.

---

Local Development & Docker

# Clone and install dependencies
npm install

# Build TypeScript
npm run build

# Start server locally on port 8080
npm start

Run via Docker:

docker build -t firecrawl-stealth-mcp .
docker run -p 8080:8080 firecrawl-stealth-mcp