March 2026 · · Copy as Markdown

Agent Trust Gateway: On-Chain Trust Scoring for AI Agents

Agents are transacting autonomously now. They're calling APIs, paying for services, and delegating tasks to other agents. But they have no standard way to check if the agent on the other end is legit before sending money.

I built the Agent Trust Gateway, a paid API that scores agent trustworthiness using on chain identity and reputation data from ERC-8004. It's live at agent-trust-gateway.port402.com and registered as Agent #21557 on Base mainnet.

The Problem

What It Does

Three core capabilities:

• Profile lookup: fetch agent identity and registration metadata from the ERC-8004 Identity Registry ($0.001)
• Trust scoring: compute a trust score from on-chain reputation data: feedback volume, ratings, identity maturity ($0.01)
• Endpoint validation: verify that an agent's declared endpoint is reachable and healthy ($0.03)

Architecture

graph TB
    Client["Client / Agent"] -->|"Request"| URL["Lambda Function URL"]
    subgraph AWS["AWS (Terraform-managed)"]
        URL --> Hono["Hono App"]
        Hono --> x402["x402 Middleware"]
        x402 -->|"402 Payment Required"| Client
        x402 -->|"Paid"| Routes["API Routes"]
    end
    Routes --> Identity["ERC-8004<br/>Identity Registry"]
    Routes --> Reputation["ERC-8004<br/>Reputation Registry"]
    Routes --> IPFS["IPFS Gateway<br/>(metadata)"]
    x402 -->|"Verify payment"| CDP["CDP Facilitator"]

• Hono framework, TypeScript, runs on AWS Lambda (arm64)
• Single esbuild bundle, Docker container, Terraform-managed infra
• x402 payment gating on all API endpoints via CDP facilitator
• Reads from two on-chain contracts: Identity Registry + Reputation Registry
• In-memory caching (10-min TTL) for registration metadata

Endpoints

Profile Lookup

GET /api/agent/1434/profile

{
  "agentId": "1434",
  "chain": "base",
  "owner": "0xF36bB95548aE036b8aDd44f94cd0D01316825a20",
  "wallet": "0xF36bB95548aE036b8aDd44f94cd0D01316825a20",
  "name": "Meerkat James",
  "description": "James is a specialized AI agent with deep expertise in robotics...",
  "image": "https://www.meerkat.town/meerkats/meerkat_019.png",
  "endpoints": [
    {
      "name": "MCP",
      "endpoint": "https://meerkat.up.railway.app/mcp/meerkat-19",
      "version": "2025-06-18"
    },
    {
      "name": "A2A",
      "endpoint": "https://meerkat.up.railway.app/agents/meerkat-19/.well-known/agent-card.json",
      "version": "0.3.0"
    }
  ],
  "supportedTrust": ["reputation", "crypto-economic", "tee-attestation"],
  "active": true,
  "registrations": [
    {
      "agentId": 1434,
      "agentRegistry": "eip155:8453:0x8004A169FB4a3325136EB29fA0ceB6D2e539a432"
    }
  ]
}

Trust Score

POST /api/agent/score/invoke

// Request
{
  "input": {
    "agentId": "1434"
  }
}

// Response
{
  "output": {
    "agentId": "1434",
    "chain": "base",
    "trustScore": 68,
    "verdict": "trusted",
    "breakdown": {
      "feedbackScore": 50,
      "identityMaturity": 60,
      "reputationConfidence": 0,
      "formula": "score = feedbackAvg + (identityMaturity * 0.3) + (reputationConfidence * 0.2)"
    },
    "feedbackSummary": {
      "count": 0,
      "averageScore": 0,
      "uniqueClients": 0
    },
    "agentName": "Meerkat James"
  }
}

Endpoint Validation

POST /api/agent/validate/invoke

// Request
{
  "input": {
    "agentId": "1434",
    "checks": ["endpoints", "wallet"]
  }
}

// Response
{
  "output": {
    "agentId": "1434",
    "chain": "base",
    "agentName": "Meerkat James",
    "endpointStatus": [
      {
        "name": "MCP",
        "endpoint": "https://meerkat.up.railway.app/mcp/meerkat-19",
        "status": "reachable",
        "latencyMs": 91
      },
      {
        "name": "A2A",
        "endpoint": "https://meerkat.up.railway.app/agents/meerkat-19/.well-known/agent-card.json",
        "status": "reachable",
        "latencyMs": 227
      }
    ],
    "walletStatus": {
      "address": "0xF36bB95548aE036b8aDd44f94cd0D01316825a20",
      "valid": true,
      "isOwner": true
    },
    "attestations": [],
    "overallVerdict": "validated",
    "issues": []
  }
}

Payment Flow

sequenceDiagram
    participant C as Client
    participant A as Agent Trust Gateway
    participant CDP as CDP Facilitator
    participant Base as Base L2

    C->>A: POST /api/agent/score/invoke
    A-->>C: 402 Payment Required
    Note over C: Sign EIP-3009<br/>(no gas)
    C->>A: Retry with X-Payment-Response
    A->>CDP: Verify + settle
    CDP->>Base: Submit transfer on-chain
    Base-->>CDP: Confirmed
    CDP-->>A: Settlement receipt
    A-->>C: 200 OK + trust score

• Gasless USDC payments on Base mainnet via EIP-3009
• Client signs off-chain, CDP facilitator handles on-chain settlement
• No gas fees for the caller

Trust Score Formula

• Feedback average from on-chain Reputation Registry
• Identity maturity bonus (how long the agent has been registered)
• Reputation confidence (volume of feedback)
• score = feedbackAvg + (identityMaturity * 0.3) + (reputationConfidence * 0.2)

A2A Protocol Support

The gateway supports Google's A2A (Agent-to-Agent) protocol, so other agents can talk to it using JSON-RPC. Natural language queries like "Get trust score for agent 42" get routed to the right endpoint automatically. It exposes three skills: profile, trust-score, and validate. All discoverable via the agent card:

{
  "name": "Agent Trust Gateway",
  "description": "ERC-8004 trust scoring and validation for the agentic economy. Query agent profiles, compute trust scores, and validate endpoints.",
  "url": "https://agent-trust-gateway.port402.com/a2a",
  "version": "0.1.0",
  "protocolVersion": "0.3.0",
  "capabilities": {
    "streaming": false,
    "pushNotifications": false
  },
  "skills": [
    { "id": "profile", "name": "Agent Profile Lookup" },
    { "id": "trust-score", "name": "Trust Score" },
    { "id": "validate", "name": "Endpoint Validation" }
  ],
  "defaultInputModes": ["text"],
  "defaultOutputModes": ["text"],
  "provider": {
    "organization": "Port402",
    "url": "https://port402.com"
  },
  "documentationUrl": "https://github.com/will-ops-agent/agent-trust-gateway",
  "entrypoints": {
    "profile": { "path": "/api/agent/:id/profile" },
    "score": { "path": "/api/agent/score/invoke" },
    "validate": { "path": "/api/agent/validate/invoke" }
  }
}

On-Chain Identity

Try It

# Query a trust score (using x402-cli)
x402 test https://agent-trust-gateway.port402.com/api/agent/score/invoke \
  --wallet $PRIVATE_KEY \
  --method POST \
  --body '{"input": {"agentId": "42", "chain": "base"}}'

Tech Stack

• Runtime: Node.js 22, AWS Lambda (arm64)
• Framework: Hono
• Blockchain: viem, erc-8004-js, agent0-sdk
• Payments: x402 (CDP facilitator, gasless USDC on Base), x402-cli
• A2A: @a2a-js/sdk
• Infra: Terraform, Docker, ECR, Secrets Manager
• Validation: Zod