This is a guardrails layer that sits in front of your AI system and validates requests using an SMT (Satisfiability Modulo Theories) solver instead of another LLM. The idea is that formal verification tools are harder to jailbreak than pattern-matching models. It runs automated reasoning checks on inputs and outputs, generates zero-knowledge proofs for auditability, and connects via streamable HTTP to api.icme.io/mcp. You'd reach for this when you need cryptographically verifiable proof that your AI constraints were actually enforced, not just that a classifier said they were. The formal methods approach means no prompt injection tricks should slip through, though you'll want to understand SMT constraints to configure it properly.
claude mcp add --transport http io.github.wyattbenno777-icme-preflight https://api.icme.io/mcp