Irmu
AI7 minIrmu Engineering

How to Give Claude, ChatGPT and Cursor Live Web Access with an MCP Server

Connect the Irmu MCP server once and your AI assistant can fetch pages, take screenshots, check Google AI Overviews and pull Booking.com prices — with setup for every client and real prompts.

Why assistants need a web tool

Claude, ChatGPT and Cursor answer from training data that is months old, and their built-in browsing gets blocked by anti-bot walls, cannot render heavy JavaScript and cannot pick a country. Ask for today's hotel prices in Lisbon or whether Google shows an AI Overview for your keyword, and you get a guess.

The Model Context Protocol (MCP) fixes this by letting the assistant call real tools. Point it at https://app.irmu.com/mcp and it gets the same infrastructure as the Irmu API: JavaScript rendering, residential IPs, geotargeting and dedicated endpoints for Google and Booking.com.

The six tools

Each tool maps to one API endpoint and uses the same credits, so there is nothing new to learn about pricing.

  • crawl — GET /crawl: any URL as HTML, markdown or AI-extracted JSON.
  • screenshot — GET /screenshot: a picture of the page, from any country.
  • google_ai_overview — GET /google-ai-overview: Google's AI answer and its cited sources.
  • booking_search — GET /booking/search: hotels and prices for a destination and dates.
  • booking_hotel — GET /booking/hotel: one property's rooms, prices and reviews.
  • crawl_poll — GET /crawl/{job}: collects slow jobs that returned a job id, called by the assistant automatically.

Connect it in one minute

In Claude Code, run the command below, then type /mcp and sign in with your Irmu account. Claude, ChatGPT, Cursor, VS Code, Windsurf and Lovable each have a slightly different screen; step-by-step instructions for all of them are on https://irmu.com/docs/mcp.

If your client lets you set headers, you can use an API key instead of signing in. Create a dedicated key for MCP so its usage shows separately in the dashboard.

oauth.sh
claude mcp add --transport http irmu https://app.irmu.com/mcp
mcp.json
{
  "mcpServers": {
    "irmu": {
      "url": "https://app.irmu.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_IRMU_API_KEY" }
    }
  }
}

Prompts that work

Be specific about the page, the country and the shape of the answer. The assistant picks the tool; you never name it.

  • "Find 4-star hotels in Lisbon for 12–14 March under €200 a night and sort them by review score."
  • "Does Google show an AI Overview for 'best web scraping api' in the US? Which domains does it cite?"
  • "Take a screenshot of our pricing page as seen from Germany and tell me if the currency is euros."
  • "Fetch https://competitor.example/pricing and give me a table of plans, prices and limits."

Slow pages may come back as a job id first. The assistant then calls crawl_poll on its own; you only see the final answer.

What it costs

Exactly what the matching endpoint costs: 1 to 25 credits for a crawl depending on rendering and IP tier, 30 flat for a Google AI Overview or a Booking.com call. There is no MCP surcharge, and the free plan works too, so you can try it before paying.

Endpoints used in this guide

Start building with Irmu today

200 free credits every month, no card required. Every API, every integration, one key.