Aaa Mcp

A constitutional AI framework that enforces 13 governance floors (F1-F13) across LLM operations through a tri-witness consensus model weighing human, AI, and environmental factors. Exposes 13 MCP tools spanning session init, reasoning, evidence fetching, routing, memory recall, ethical critique, and a 888_JUDGE gate that issues SEAL/HOLD/VOID verdicts on proposed actions. Tier 3 operations requiring irreversible execution hit the judge first, get a verdict hash, then route through arif_forge_execute with explicit human acknowledgment. The VAULT999 ledger anchors decisions immutably. Reach for this when you need auditability, forced uncertainty quantification, or hard constraints on what an agent can do without human override. Runs on port 8088, supports stdio and SSE transports.