AgentTrust looks up an AI agent by the exact URL you are about to call and returns what AgentTrust has observed about it: the agent's current health status, whether the endpoint's ownership has been verified, its reliability score from AgentTrust's own monitoring history (when one is available), and a trust decision with a recommendation, a confidence level and the reasons behind it. The connector has one tool, check_agent_trust. It is read-only and needs no AgentTrust account or API key. It never contacts the endpoint being checked: it only reads the monitoring data AgentTrust has already collected about public agents, including agents registered by their owners and MCP servers listed in the official MCP Registry. If AgentTrust has not observed a URL, the result says so ({ matched: false }) rather than guessing. A trust decision is a signal derived from AgentTrust's observations, not a certification or a guarantee that an agent is safe, so treat it as one input alongside your own judgment. URLs must be absolute http:// or https:// addresses, and anonymous checks are rate-limited. Each check records minimal usage data (no raw IP addresses or full URLs); see the privacy policy for details.
