Report privately through GitHub's security advisories rather than a public issue. You'll get a reply within a week.

This matters more than usual here, because the tool asks you for keys to three paid APIs. Every claim below is checkable in the source — the whole thing is a few hundred lines, and src/engines.ts is the only file that touches a key.

• Keys are read from environment variables, at call time. OPENAI_API_KEY, GOOGLE_API_KEY, PERPLEXITY_API_KEY. Nothing is read at import time, so importing the library or running the MCP server never touches your environment unless you actually run an audit.
• Each key is sent to that engine's own API and nowhere else. The complete list of hosts this software will ever contact: api.openai.com, generativelanguage.googleapis.com, api.perplexity.ai, and the site you asked it to audit.
• Keys travel in headers, never in URLs. Including Google's, which its docs permit as a ?key= query parameter — URLs end up in proxy logs, error messages and shell history in a way headers don't.
• Nothing is stored or transmitted anywhere else. No telemetry, no analytics, no crash reporting, no phone-home, no account, no server component. There is nothing to opt out of.
• Keys are never written to output. Not in --json, not in error messages.

Two runtime dependencies, both only used by the MCP server: @modelcontextprotocol/sdk and zod.

The readiness audit requests, over plain HTTPS with no authentication:

• the homepage,
• /robots.txt, /llms.txt, /sitemap.xml,
• a HEAD on up to 10 same-origin links found on the homepage.

It identifies itself honestly as crazyseo/1.0 (+https://github.com/d08m/crazyseo) and does not retry with a spoofed browser user-agent — that was measured, didn't work, and is documented in src/readiness.ts.

It respects nothing about robots.txt beyond reading it, because it fetches only your own site's public homepage and three well-known files — roughly what a browser does when you visit. It is not a crawler and has no queue.

Audit whatever you own or have permission to audit. Pointing it at a site to measure that site's AI visibility is the intended use; that's public information either way. Rate-limiting yourself is your responsibility.