Agent Policy Enforcement Mcp

This gives you IAM-style policy enforcement between agent pairs in multi-agent systems. You define rules like "orchestrator may only call billing agent when amount is under 1000" and gate every agent-to-agent call through an evaluate_call check. It logs decisions with signed attestations, which is useful if you're tracking EU AI Act Article 14 audit trails or ISO 42001 compliance. The core is really just policy definition, permission inheritance, and runtime evaluation. Reach for this when you're building agent orchestration and need to restrict which agents can invoke which other agents under what conditions, rather than letting everything talk to everything.