Call for a human OK before an irreversible action: approve link + Ed25519 receipt
Scored across 3 tools
The three tools map to distinct phases: request_approval creates, get_approval polls status, verify_receipt validates a JWS. The only potential confusion is that get_approval already returns the signed receipt, so an agent might skip verify_receipt; the descriptions make the separation discoverable but not effortless.
All three follow a clean snake_case verb_noun pattern (get_approval, request_approval, verify_receipt) with consistent use of the 'approval/receipt' domain nouns. No mixed conventions or vague verbs.
Three tools is lean but each covers a genuinely distinct step in the approval lifecycle, so nothing feels redundant. It sits just under the 'well-scoped 3-15' ideal, with room for one or two lifecycle helpers before it would feel heavy.
The core loop (request -> poll/wait -> verify receipt) is fully covered, including long-polling and x402 payment handling. Gaps exist around lifecycle management—no cancel/revoke of a pending request and no list of outstanding approvals—which could strand an agent that needs to withdraw a request.
Free. Status of an approval request (pending | approved | denied | expired). When decided, includes the Ed25519-signed receipt (compact JWS). Optional wait_seconds long-polls until a decision.
Inputs: approval_id (required), wait_seconds.
Create a human approval request for an irreversible action. Returns approve_url (pass it to the responsible human through your own channel) and approval_id. Optional wait_seconds keeps the call open (progress notifications every 10 s when the client sends a progressToken) until the human decides. 3 free per day, then 0.005 USDC via x402 (params._meta["x402/payment"]).
Inputs: action_summary (required), risk, requester, callback_url, +2 more.
Free. Verify an Approval Relay decision receipt (compact JWS) against the current public key and return its decoded payload.
Inputs: receipt_jws (required).
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Human approval for irreversible AI agent actions, bound to the exact tool call by a passkey tap
Human-in-the-loop approval for agent actions, with verifiable action-bound receipts.
Ask a human for legal review, confirmation, a signature, or a physical-world act
Preventive human-approval write-gate for AI agents: writes commit only after a human approves.
Add one secure layer between your agents and this server.
