MCP/A2A detection / Runtime interception
MCP/A2A detection

Runtime interception, at both levels

SentinelleIA sits between your agents and their tools, and between your agents themselves, at runtime.

Submit my system

In production, not in a lab

Runtime detection, deployable on-premise via Docker. The equivalent of an EDR, but for agentic AI: we observe real behavior and block fail-closed.

What we intercept

MCP (agent to tool) and A2A (agent to agent) together. Chained audit log, 25 MCP/A2A audit event types, TOFU pinning, verified delegation chain, fail-closed behavior.

Compliance

Explicit mapping to Articles 9 to 15 (rules, chained audit log, reports, model cards). Industrializing this into end-to-end automatic generation is the object of TRL 6 to 8 maturation.

See the detectors

Four threat families, detected at runtime.

See the detectors