Agentic Proof ProtocolVerifiable authority for AI agents
Monad 10143
MONAD · TRUST, IDENTITY & AI INFRASTRUCTURE

The model proposes.
The protocol authorizes.

Give autonomous agents scoped authority, record tamper-evident action commitments, and let any application independently verify what happened without trusting the agent operator.

AI proposal
→
Policy gate
→
Scoped authority
→
Proof onchain
REAL TESTNET TRANSACTIONS · NO WALLET REQUIRED

Inspect the recorded $361 agent action.

See the original payload match its onchain commitment, then change $361 to $399 and watch the proof fail. This authorization has since been revoked: the historical action remains valid, but new actions are blocked.

Agent serviceOFFLINEWaiting for API
Deterministic policyUNKNOWNNo policy loaded
Audit integrity—Local SHA-256 hash chain
Independent verifierDIRECT RPC0xC8E9E…54010
REFERENCE FLOW

Travel agent · scoped booking authority

BOOK_TRAVEL ≤ $400 · 30 min
Live protocol workflow
01
Register the reference agentThe agent wallet binds its own address to an onchain Agent ID.
02
Grant scoped authorityUser wallet → BOOK_TRAVEL · scope 0x73753…fac6a · max $400.
03
AI agent proposes an itineraryThe model may choose an offer, but it cannot grant authority or bypass deterministic policy.
04
Record agent actionThe deterministic policy must return ALLOW before the agent signer can submit.
05
Negative proof: try $401Hard cap must fail closed before chain execution.
INDEPENDENT VERIFIER

Verify without trusting the agent backend

Verification result
No proof loaded yet.
PORTABLE PRIMITIVES

One protocol, more than one agent

Same SDK · same registry
Reference integration #1Travel Agent

BOOK_TRAVEL · Singapore scope · max $400 · proposed $361.

Reference integration #2Procurement Agent

PURCHASE_ITEM · approved vendor scope · max $200 · proposed $120.

CONNECT AN EXTERNAL AI AGENT

MCP privilege boundary

Claude, Codex or another MCP client can inspect status, evaluate proposals, prepare commitments and verify proofs. The MCP surface cannot grant authority, revoke a user capability or submit an onchain action.

bun run dev:mcpstdio · API → http://127.0.0.1:8787
get_statusevaluate_actionprepare_actionverify_action
Agentproposes
→
Control PlaneALLOW / BLOCK / ESCALATE
→
Monad Registryauthority + commitments
→
Verifierdirect RPC
Ready. Start by registering the reference agent and connecting your wallet.