Give your AI agent the web: connecting an MCP server in one config block

Agents are only as useful as what they can reach. Model Context Protocol (MCP) has become the standard way to hand them new capabilities — and web access is the highest-leverage one you can add. Here's how to connect a hosted scraping MCP in about a minute.

What MCP gives an agent

Without web access, an LLM answers from training data. With an MCP server exposing tools like scrape, crawl, and search, the same agent can ground answers in live pages: read the docs you point it at, check a competitor's pricing page, summarize a thread, verify a claim.

The one-block install

Add this to your MCP client config (Claude Desktop, Cursor, Codex, Hermes — they all speak the same format):

{
  "mcpServers": {
    "fastcrawl": {
      "url": "https://mcp.fastcrawl.net/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Grab a free key (2,000 credits/month, no card), paste the block, restart your client. Twelve tools appear: scrape, crawl, map, search, extract, screenshot, pdf, parse, plus monitor management.

Prompts worth trying on day one

• "Scrape this changelog and summarize what shipped this week."
• "Map example.com, then tell me which sections exist."
• "Search for the latest on X and cite sources."
• "Watch this pricing page daily and tell me when it changes." (monitors)

Why hosted beats local for web tools

You could run Playwright locally — and for a hobby script, you should. But local browser fleets mean version drift, memory pressure, anti-bot whack-a-mole, and zero visibility into failures. A hosted engine renders JS at the edge, caches repeats, detects bot walls, and bills flatly per page. Your agent stays light; the heavy lifting happens where browsers are cheap.