Agent Utility Services (AUS) Gateway

An enterprise-hardened API gateway designed for autonomous AI agent networks. AUS provides 8 specialized utility engines exposed through a Model Context Protocol (MCP) manifest and monetized via HTTP 402 payment handshakes on Base.

---

🏛️ Architecture Overview

[ External AI Agent ]
         │
         ▼  (HTTPS / HTTP/2)
[ Cloudflare Tunnel Edge ]
         │
         ▼  (Port 3000)
[ AUS Express Gateway ]
   ├── Rate Limiter (100 req / 60s)
   ├── 402 Payment Negotiation & Anti-Replay Guard (In-Memory / Hash-Lock)
   └── Route Dispatcher
         ├── /mcp.json (Discovery Manifest)
         └── /tools/* (8 Specialized Utility Engines)

---

🚀 Quick Start

1. Prerequisites

• Node.js (v20+)
• cloudflared CLI
• Valid .env configuration:

  PORT=3000
  NODE_ENV=production
  ANTHROPIC_API_KEY=your_key_here
  TREASURY_BASE_ADDRESS=0xYourBaseTreasuryAddress

2. Start the Server & Tunnel

# Terminal 1: Launch AUS Gateway
node server.js

# Terminal 2: Expose via Cloudflare Tunnel
cloudflared tunnel --protocol http2 --url http://127.0.0.1:3000

---

🛠️ MCP Tool Manifest (GET /mcp.json)

Agents discover all endpoints dynamically by querying /mcp.json:

{
  "mcpVersion": "1.0.0",
  "serverInfo": {
    "name": "agent-utility-services",
    "version": "2.1.0",
    "tier": "Enterprise-Hardened"
  },
  "tools": [
    { "name": "schema_sanitizer", "endpoint": "/tools/schema-sanitizer", "price": "0.10 USDC" },
    { "name": "financial_audit", "endpoint": "/tools/financial-audit", "price": "0.20 USDC" },
    { "name": "verification_oracle", "endpoint": "/tools/verification-oracle", "price": "0.05 USDC" },
    { "name": "sandbox_execution", "endpoint": "/tools/sandbox-execution", "price": "0.25 USDC" },
    { "name": "media_transcoder", "endpoint": "/tools/media-transcoder", "price": "0.15 USDC" },
    { "name": "geospatial_verifier", "endpoint": "/tools/geospatial-verifier", "price": "0.08 USDC" },
    { "name": "claude_reason", "endpoint": "/tools/claude-reason", "price": "0.12 USDC" },
    { "name": "agentic_audit", "endpoint": "/tools/agentic-audit", "price": "0.25 USDC" }
  ]
}

---

⚡ 402 Payment Protocol & Anti-Replay

Handshake Flow

1. Unsettled Request: The agent calls a tool endpoint without the x-payment-receipt header.
2. 402 Challenge: The gateway halts execution and returns HTTP 402 Payment Required:

   {
     "status": 402,
     "error": "Payment Required",
     "payment_request": {
       "token": "USDC",
       "network": "Base",
       "amount": "0.12",
       "destination_address": "0xYourBaseTreasuryAddress"
     }
   }

3. Settled Execution: The agent executes the on-chain transfer and repeats the call with the header:

   x-payment-receipt: 0xBaseTx_f7a9...

4. Anti-Replay Burning: The gateway consumes and registers the hash. Any subsequent call with the identical receipt returns HTTP 409 Conflict.

---

📦 Verified Utility Fleet Reference

1. Claude Reasoning (/tools/claude-reason)

Deep analytical reasoning powered by Claude Opus 5.

• Payload:

  {
    "system_instruction": "You are a treasury analyst.",
    "prompt": "State 2 benefits of agent micro-settlement channels."
  }

2. Sandbox Execution (/tools/sandbox-execution)

Secure, isolated Node.js vm sandbox with configurable timeout ceilings.

• Payload:

  {
    "code": "const sum = input.items.reduce((a, b) => a + b, 0); output = { total: sum };",
    "context_data": { "items": [100, 200, 300] },
    "timeout_ms": 1500
  }

3. Verification Oracle (/tools/verification-oracle)

Deterministic SHA-256 state attestation and integrity hashing.

• Payload:

  { "data": "Base is an Ethereum Layer 2 incubated by Coinbase." }

4. Media Transcoder (/tools/media-transcoder)

MIME analysis, byte size verification, and base64/URI media inspection.

• Payload:

  {
    "data_uri": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAE...",
    "target_format": "webp"
  }

5. Financial Audit (/tools/financial-audit)

Batch transaction ledger reconciliation, volume computation, and anomaly flagging.

• Payload:

  {
    "transactions": [
      { "id": "tx_01", "amount": 250.00, "type": "credit", "category": "api_revenue" },
      { "id": "tx_02", "amount": 35.50, "type": "debit", "category": "gas_overhead" }
    ]
  }

6. Schema Sanitizer (/tools/schema-sanitizer)

XSS/HTML stripping, key normalization, and sensitive field masking.

• Payload:

  {
    "data": { "user": "agent_01", "api_key": "sk_live_9999", "bio": "<script>alert(1)</script>hello" },
    "rules": { "stripHtml": true, "maskFields": ["api_key"] }
  }

7. Geospatial Verifier (/tools/geospatial-verifier)

Haversine distance calculation and geofence boundary verification.

• Payload:

  {
    "origin": { "lat": 52.2681, "lng": -113.8112 },
    "destination": { "lat": 51.0447, "lng": -114.0719 }
  }

8. Agentic Audit (/tools/agentic-audit)

Execution step health scoring, latency tracking, and circular loop detection for autonomous workflows.

• Payload:

  {
    "agent_id": "agent_treasury_01",
    "steps": [
      { "step": 1, "action": "fetch_balance", "status": "success", "latency_ms": 45 },
      { "step": 2, "action": "burn_receipt", "status": "success", "latency_ms": 120 }
    ]
  }

---

🛡️ Security Hardening

• Rate Limiting: express-rate-limit enforces a ceiling of 100 requests per 60-second window per IP.
• Receipt Collision Guard: In-memory ledger flags and blocks replayed transaction hashes immediately.
• VM Sandboxing: Script timeouts prevent CPU exhaustion attacks; execution contexts lack access to host process, fs, or network bindings.
• Edge Routing: Direct origin isolation via Cloudflare Tunnel over HTTP/2.