What can your agent
actually do?
Paste your tool definitions. See how many have zero policy coverage, the exfiltration paths hiding in the combination, and a starter policy that closes them.
Scanning a live MCP server?
Run this against your server and paste the output. We don't take URLs here - fetching arbitrary hosts server-side is an SSRF vector.
curl -s -X POST $MCP_URL -H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Answering the AI section of a security review
Vendor questionnaires and SOC 2 audits increasingly ask a version of the same question: what can your AI agents actually do, and what stops them from doing something they shouldn't? "We reviewed the prompt" isn't an answer a security team accepts anymore.
This scan gives you a concrete starting point: which of your agent's tool calls have a policy attached, which don't, and a real, installable policy that closes the gap - something you can point to, not just describe.
If you need a dated, exportable record of enforcement decisions over time (not just a snapshot of your tool schema) rather than declared intent, that's what Comply reports are for once you're enforcing policies in production.