Problem. Agents have no transparent checkpoint that inspects traffic for
prompt injection and guardrail-bypass before it reaches the agent brain.
Target user. Any operator who wants every agent request screened in one place.
Market gap. Lakera, Rebuff, Prompt Security are all "call our API" — nobody
ships a transparent MCP-native traffic firewall (Scout research, 2026-05-16).
MVP scope (shipped). A checkpoint dashboard: incoming/outgoing traffic
logged, scanned for injection, blocked/allowed verdicts, encryption-on-pass.
Acceptance criteria. (1) Traffic rows show blocked vs allowed with reasons.
(2) An injection attempt is visibly caught. (3) TLS/encryption status shown.
(4) Mobile-clean, loads < 2s.
Out of scope (v1). Real MCP middleware interception; live packet capture.
Next. Phase 2 — MCP middleware layer (regex + classifier) + traffic log store.