The connector links an MCP client to your Onyx deployment. Policies, requests, decisions, certificates, and the audit trail all live on your infrastructure — nothing routes through The Onyx Foundry. See the privacy policy.
Gating tools decide an action and append a record to the hash-chained audit trail; verification and audit tools are read-only.
| Tool | Does | Access |
|---|---|---|
| gate_action | Decide one authorization request against your policy set — fail-closed, with the decision proved. | writes trail |
| gate_tool_call | Decide whether an agent tool call may proceed — the runtime gate for agent stacks. | writes trail |
| gate_transaction | Decide a custody-style transaction (ordered rules, first match, quorum approvals) with the quorum report. | writes trail |
| verify_policies | Verify assert/obligate properties over all requests — counterexamples on violation, certificates on pass. | read-only |
| verify_tx_policy | Verify custody-policy safety properties (e.g. no single signer) with witness transactions and a certificate bundle. | read-only |
| verify_claim | Prove or refute premises ⊢ conjecture; a non-theorem returns a concrete countermodel. | read-only |
| check_smt | Decide satisfiability over the built-in theories; every unsatisfiable answer carries a re-checkable certificate. | read-only |
| audit_proof | Step-by-step audit of an external prover's proof (TSTP or Alethe format), verdict per step. | read-only |
The connector reads (verifications, proofs, audits) and writes (audit-trail records from gating decisions) — entirely on your server. Onyx decides whether actions and transactions comply with your declared policy; it never initiates or executes transactions or moves funds.
contact@onyxfoundry.ai · security reports: security.txt