io.github.wyattbenno777/icme-preflight

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Jailbreak-proof AI guardrails. Automated Reasoning SMT solver, not an LLM. ZK proofs included.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "icme-preflight": {
      "url": "https://api.icme.io/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (14)

• check_logic

  Catch contradictions in reasoning before acting on it. FREE — no account needed. Extracts quantitative and logical claims from any plan, calculation, or chain of thought, then uses a Z3 SAT solver to mathematically prove whether they contradict each other. This is formal verification, not an LLM second-guessing itself. Returns CONSISTENT, CONTRADICTION, or UNKNOWN with the extracted claims.

  Parameters:

  • reasoning (required) — string — The reasoning, plan, or chain of thought to check. Be specific — include numbers, conditions, and constraints for the best results (max 2000 chars)
  • show_smt — boolean — Include the generated SMT-LIB2 formula in the response for inspection (default: false)

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• check_relevance

  Free relevance screen — checks whether an action touches any of your policy variables before running a paid check. Returns a relevance score and which variables matched. If should_check is true, run check_action. If false, the action is unrelated to your policy — skip the paid check. No credits charged. Requires api_key.

  Parameters:

  • policy_id (required) — string — Policy ID from make_rules
  • action (required) — string — The agent action to screen (max 2000 chars)
  • api_key (required) — string — Your ICME API key
  • threshold — number — Relevance threshold (0.0 to 1.0). Default 0.0 — any match triggers should_check. Optional.

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• make_rules

  Turn a plain-English policy into a jailbreak-proof guardrail. Write rules the way you would explain them to a colleague — ICME compiles them into formal logic (SMT-LIB) that a mathematical solver enforces. No prompt engineering. No LLM judges. The solver either proves an action complies or it doesn't. Returns a policy_id and test scenarios for review. Costs 300 credits. Requires api_key.

  Parameters:

  • policy (required) — string — Your policy in plain English, e.g. 'No action may access user data without explicit consent. External API calls require approval above $100.' Up to 50 rules.
  • api_key (required) — string — Your ICME API key (from create_account or create_account_card)

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• check_action

  Enforce a guardrail: verify an agent action against a compiled policy using formal verification. An SMT solver — not an LLM — determines whether the action satisfies every rule. Returns SAT (allowed) or UNSAT (blocked) with extracted values and a cryptographic ZK proof that the check was performed correctly. Cannot be jailbroken. 1 credit ($0.01). Requires api_key. Tip: end the action with an explicit claim like 'I assert this complies with the policy' for best extraction.

  Parameters:

  • policy_id (required) — string — Policy ID from make_rules
  • action (required) — string — The agent action to verify against the policy (max 2000 chars)
  • api_key (required) — string — Your ICME API key

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• check_action_paid

  Same formal verification as check_action, but pay per call with x402 ($0.10 USDC on Base) instead of using credits. No API key or account needed — any agent with a wallet can verify actions on the fly. Returns SAT (allowed) or UNSAT (blocked) with extracted values and optional ZK proof.

  Parameters:

  • policy_id (required) — string — Policy ID from make_rules
  • action (required) — string — The agent action to verify against the policy (max 2000 chars)

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• quick_check

  Fast yes/no guardrail verdict — returns only ALLOWED or BLOCKED with no proof details. Use when you need a lightweight gate and don't need the full SAT/UNSAT report or ZK proof. Same formal verification under the hood, just a minimal response. 1 credit. Requires api_key.

  Parameters:

  • policy_id (required) — string — Policy ID from make_rules
  • action (required) — string — The agent action to check (max 2000 chars)
  • api_key (required) — string — Your ICME API key

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• verify_proof

  Independently verify a ZK proof from a prior check_action call. Confirms the guardrail check was performed correctly without re-running it — any third party or monitoring agent can verify in under one second. No additional cost. Wait a few minutes after the check for the proof to be generated. Single-use per proof.

  Parameters:

  • proof_id (required) — string — The proof_id returned by a prior check_action or check_action_paid call

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• get_scenarios

  Review auto-generated test scenarios for a compiled policy. Shows example actions that should pass and fail, so you can verify the policy behaves as intended before deploying it. Requires api_key.

  Parameters:

  • policy_id (required) — string — Policy ID from make_rules
  • api_key (required) — string — Your ICME API key

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• run_tests

  Run saved test cases against a policy to confirm it blocks what it should block and allows what it should allow. Run this after make_rules and before using the policy in production. Requires api_key.

  Parameters:

  • policy_id (required) — string — Policy ID to test
  • api_key (required) — string — Your ICME API key

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• create_account

  Create an ICME Preflight account with x402 USDC payment ($5.00 on Base). Returns an API key and 325 starting credits immediately. Save the api_key — it is shown only once. Use create_account_card instead if paying by credit card.

  Parameters:

  • username (required) — string — Unique username (1-32 chars, alphanumeric + hyphens/underscores)

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• create_account_card

  Create an ICME Preflight account with a credit card ($5.00 via Stripe). Returns a checkout_url — open it in a browser to pay. After payment completes, call poll_session with the session_id to retrieve your API key.

  Parameters:

  • username (required) — string — Unique username (1-32 chars, alphanumeric + hyphens/underscores)

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• top_up

  Add 500 credits to your account via x402 USDC payment ($5.00 on Base). Use top_up_card instead for credit card payment with volume discounts. Requires api_key.

  Parameters:

  • api_key (required) — string — Your ICME API key

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• top_up_card

  Add credits via credit card with volume bonuses. Returns a checkout_url. Tiers: $5 = 500 credits, $10 = 1,050 (+5%), $25 = 2,750 (+10%), $50 = 5,750 (+15%), $100 = 12,000 (+20%). Credits never expire. Requires api_key.

  Parameters:

  • amount_usd (required) — integer — Top-up amount in USD: 5, 10, 25, 50, or 100
  • api_key (required) — string — Your ICME API key

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

• poll_session

  Check the status of a Stripe card payment. Returns pending or complete. On completion after signup, returns the api_key. Call this after create_account_card or top_up_card once the user has paid.

  Parameters:

  • session_id (required) — string — The session_id returned by create_account_card or top_up_card

  Hints: read-only not declared · destructive not declared · idempotent not declared · open-world not declared

Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.