MCP/A2A detection

MCP and A2A: the blind spots of classic security

Your firewalls and API gateways do not see agent-to-agent traffic. That is where attacks propagate.

Submit my system

Two traffic planes

↕️

MCP (north-south)

Traffic between an agent and its tools (Model Context Protocol). The path by which an agent reads, writes and calls real systems.

↔️

A2A (east-west)

Traffic between agents (Agent-to-Agent). Rarer to monitor, and more dangerous: it is the channel for the cascading propagation of a compromise.

Why it is blind

WAFs and API gateways inspect classic application north-south traffic. They understand neither delegation between agents nor content injected indirectly through tools. The A2A-specific risk, cascading propagation, escapes them entirely.

See how we intercept

SentinelleIA sits at both levels, at runtime.

See interception