Agent quickstart

Set up the Prefetch API (https://prefetch.io) in this project.

Docs index: https://docs.prefetch.io/llms.txt (full text: https://docs.prefetch.io/llms-full.txt).
Read only the pages you need.

1. Ask me for my Prefetch API key. I get it from https://prefetch.io/dashboard.
   Never print it, log it, or commit it.
2. Store it as PREFETCH_API_KEY: in .env for application code (make sure .env is in .gitignore),
   or in this agent's MCP config if I only want to use Prefetch from chat.
3. Ask me which I want, then do it:
   a. MCP for this agent: add the remote MCP server https://api.prefetch.io/mcp
      (Claude Code: `claude mcp add --transport http prefetch https://api.prefetch.io/mcp`,
      then authenticate from /mcp; Cursor: add it to .cursor/mcp.json).
   b. Application code: call the REST API with the X-API-Key header. There is no SDK; use the
      project's existing HTTP client. Every response is {success, data, meta}; on failure branch
      on error_code and retry only when retryable is true.
4. Verify: call GET https://api.prefetch.io/classify?url=https://stripe.com (5 credits) and
   show me data.classification.category.
5. Tell me what you changed, and link the docs pages for the endpoints I'm likely to use.