@metalift/mcp

Metalift MCP server for AI agents. Exposes scrape, crawl, map, and batch tools over stdio for Cursor, Claude Desktop, and other MCP clients.

What is Metalift?

Metalift Cloud is a hosted web context platform — scrape, crawl, and map sites into LLM-ready markdown for AI agents. This MCP server connects your AI client to the Metalift API.

Related MCP server: Crawl4AI RAG MCP Server

Get an API key

1. Sign up for Metalift Cloud and verify your email.

2. Complete onboarding and copy your API key from Dashboard → API keys.

New accounts receive 1,000 free credits/month. Set METALIFT_API_KEY in your MCP config (see below). Do not commit keys to git.

Install

Published on npm as @metalift/mcp. Default config uses npx:

{
  "mcpServers": {
    "metalift": {
      "command": "npx",
      "args": ["-y", "@metalift/mcp"],
      "env": {
        "METALIFT_API_URL": "https://api.metalift.ai",
        "METALIFT_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Claude Desktop: paste the same JSON into claude_desktop_config.json (Settings → Developer → Edit Config). See MCP setup — Claude Desktop and examples/claude-mcp.json.

Corporate Windows / SSL inspection: if npx fails with UNABLE_TO_VERIFY_LEAF_SIGNATURE, install locally and use node — see examples/claude-mcp-local.json / cursor-mcp-local.json and MCP setup troubleshooting.

Troubleshooting

Full guide: packages/platform-web/docs/mcp-setup.md.

Web search vs scrape

Search and scrape are separate tools with separate billing:

1. metalift_web_search — returns up to 10 search snippets (2 credits, English by default). Review titles and snippets first.

2. metalift_scrape — fetches page content for URLs you choose (1+ credits per URL). Default response_detail=compact (~16k chars, no link lists). Use standard for full articles or full for raw JSON with all links.

Do not auto-scrape every search result. See Web search for agent workflow examples.

Environment variables

Tools