Agent Perimeter
Scans the tools an AI agent can reach and proves — not guesses — which ones an attacker could hijack.
It flags suspicious text, then drives a live agent through the injection path and records demonstrated impact. And every published number is regenerable by someone who doesn’t trust it.
- MCP server
- Discovery
- Capability graph
- 34 checks · 8 groups
- Findings + reproductions
- SARIF / HTML / CI drift gate