Every action verified. Every agent scored. All of it on-chain.
The infrastructure layer that should sit between autonomous agents and real capital doesn't exist yet.
AI agents are beginning to move real capital.
Trading agents executing swaps. Yield agents managing lending positions. Portfolio agents rebalancing across protocols. Strategy agents taking leveraged positions.
They're autonomous by design. That's what makes them powerful.
An agent turns your instruction into a raw transaction — bytes on a chain you can't read, routed through contracts you haven't audited, settling in a world that doesn't forgive mistakes.
That's the gap AgentScan fills.
intent.commit()
You give your agent a goal.
agent.propose()
Your agent constructs the transaction. Before any funds move, it submits the proposal to AgentScan.
agentscan.verify()
AgentScan evaluates the proposal — against your original intent, against real chain state, against every adversarial signal in the network. A verdict is issued. Reasoning is recorded.
escrow.settle()
Verified: the escrow releases. Not verified: nothing moves.
The agent never holds unilateral authority. Neither does AgentScan. Secured through multi-party computation.
Most checks ask: is this transaction valid? AgentScan asks: does it do what you actually asked for?
A routing contract can expose a clean interface while its underlying logic tells a different story. Verification catches the divergence — not by reading the label, but by evaluating the outcome.
Five individually reasonable transactions can compose into something harmful. AgentScan evaluates intent across the full sequence, not each step in isolation.
An agent can interact with a contract that launched yesterday. No prior integration required. The outcome is what gets evaluated — not the protocol’s reputation.
Connect the hosted MCP server. Every transaction your agent proposes passes through AgentScan before it reaches a signer. APPROVE or REJECT — plus a full reasoning trace your agent can act on.
Works with Claude Code, Cursor, LangChain, CrewAI, OpenAI Agents, custom builds.
One curl. Get your API key and MPC escrow wallet on Base Sepolia.
$ curl -s -X POST \ agentscan-api.../auth/register \ -H "Content-Type: application/json" \ -d '{}'
// Response { "apiKey": "as_live_0ec1...", "escrowAddress": "0xFDf7..." }
Add AgentScan to your MCP config. Claude Code, Cursor, or any MCP client.
{
"mcpServers": {
"agentscan": {
"url": "https://agentscan-mcp-
...run.app/mcp",
"headers": {
"Authorization":
"Bearer as_live_..."
}
}
}
}Ask your agent to verify a transaction. AgentScan simulates it and returns APPROVED or REJECTED with full reasoning.
> Verify swapping 2 WETH for USDC on Uniswap // AgentScan response { "decision": "APPROVED", "reasoning": "2 WETH swapped for USDC, no unexpected transfers." }
getUserAddress
Fetch your escrow address and chain info.
requestAndAwaitVerification
Submit a transaction for full pipeline verification. Returns decision, reasoning, and simulation trace.
checkVerificationStatus
Look up an earlier verification by request ID.
Every agent entering the network gets continuously probed with adversarial scenarios. Every test result feeds collective knowledge. Every verdict sharpens the reputation signal for every other agent.
Trust isn't a static check. It's a living record that compounds over time. Honest agents earn it. Dishonest agents don't.
Every verdict. Every score. Every test. On-chain.
Marketplaces rank agents by score. Wallets surface the safest ones first. Institutions run due diligence before delegating capital. All from the same public data layer.
AgentScan's verification pipeline combines deterministic analysis, intent-execution alignment, and adversarial multi-node consensus. Read the paper →
From the team behind Zeru and Z-Score — building on-chain reputation infrastructure since 2024.