1. Executive Summary
As decentralized organizations (DAOs) and AI agents become dominant economic actors, the traditional banking stack is becoming obsolete. Protocol Banks introduces a programmable, cross-chain treasury management layer designed for the future of work. By abstracting chain-specific complexities and integrating standard accounting practices directly with on-chain events, Protocol Banks enables seamless financial operations for the next generation of digital enterprises.
In v2.0, Protocol Banks evolves from a payment platform into AI-native payment infrastructure — any AI agent can discover our services (ERC-8004), authenticate with on-chain identity (SIWE), communicate via signed messages (A2A Protocol), and call payment tools directly (MCP Server).
2. The “x402” Protocol
Gasless Enterprise Settlements
Protocol Banks leverages the x402 Protocol (based on ERC-3009) to separate payment authorization from execution.
- Enables “CFO Approval” workflows where the approver doesn't need ETH/Gas.
- Allows AI Agents to propose payments securely via EIP-712 signatures.
- Facilitates recurring billing and subscriptions on-chain.
- Enables machine-to-machine micropayments via HTTP 402 status codes.
3. Market Analysis
Modern Web3 finance teams face a “fragmentation trilemma” that hinders adoption:
Chain Silos
Assets split across EVM, Solana, TRON, and Bitcoin layers.
Data Blindness
Explorers show hashes, not “Payroll” or “Vendor” context.
Agent Gap
AI Agents lack standardized ways to discover, authenticate with, and call payment services.
4. Product Architecture
Protocol Banks acts as a non-custodial overlay. We do not hold funds; we orchestrate them.
- Unified Batch Engine: Smart routing logic that bundles transactions to minimize gas fees and administrative time. Go payout-engine for 500+ TPS throughput.
- Local-First Privacy: “Wallet Tags” and financial metadata are encrypted locally or via RLS policies, ensuring your supplier list remains your trade secret.
- Agent-Ready APIs: Full REST API with API key authentication, budget management, payment proposals, auto-execute rules, and webhook notifications.
- Multi-Chain Support: 9 blockchains (Ethereum, Polygon, Arbitrum, Base, Optimism, BNB Chain, Solana, Bitcoin, TRON) with unified address detection and network-specific optimization.
5. AI-Native Architecture
Protocol Banks v2.0 is designed from the ground up as AI-native payment infrastructure. Any AI agent — whether a Claude MCP client, an autonomous trading bot, or a DAO treasury manager — can interact with our payment services through open, standardized protocols.
ERC-8004 Agent Card
Decentralized identity for AI agents. DID-based discovery via /.well-known/agent.json. EIP-191 signed cards prove agent authenticity on-chain.
SIWE + JWT Authentication
AI agents authenticate using Sign-In with Ethereum (EIP-4361). Private key signs a challenge, receives a 1-hour JWT access token and 30-day refresh token. No browser required.
A2A Protocol
Agent-to-Agent communication via JSON-RPC 2.0. Every message carries an EIP-191 signature with nonce-based replay protection and a 5-minute timestamp window.
MCP Server
Model Context Protocol server exposes 8 payment tools. Claude, GPT, and other AI models call tools directly via Streamable HTTP or stdio transport.
6. AI Agent Authentication
Protocol Banks implements a three-step SIWE (Sign-In with Ethereum) authentication flow that allows AI agents to log in autonomously using only their private key:
Request Nonce
GET /api/auth/siwe/nonce — Server generates a single-use challenge nonce (5-min expiry). The nonce can only be used once to prevent replay attacks.
Sign & Verify
POST /api/auth/siwe/verify — Agent constructs an EIP-4361 message, signs it with their private key, and submits both. Server verifies the signature, consumes the nonce, and returns a JWT access token (1h) plus a refresh token (30d).
Auto-Refresh
POST /api/auth/siwe/refresh — Before the access token expires, the agent submits the refresh token to get a new JWT. Sessions persist up to 30 days without re-authentication.
7. Agent-to-Agent Communication
The A2A Protocol enables standardized, cryptographically secured communication between AI agents. External agents can request payments, get quotes, and track transactions — all through signed JSON-RPC 2.0 messages.
A2A Message Flow
- Discovery: Resolve agent DID via
/.well-known/agent.jsonor/api/agents/cards/resolve - Handshake: Exchange capabilities and supported protocols
- Request Payment: Agent sends a signed payment request with amount, token, and recipient
- Get Quote: Server returns a fee estimate and execution plan
- Confirm & Track: Submit on-chain transaction and track status via task ID
Security guarantees: Every A2A message is verified with EIP-191 signature recovery. Nonces are stored in the database with a unique constraint to prevent replay. Messages older than 5 minutes are rejected. All interactions are logged to the audit trail.
8. MCP Server Integration
The Model Context Protocol (MCP) server allows any compatible AI model — Claude, GPT, or custom agents — to directly invoke payment tools as native function calls. This eliminates the need for wrapper code or API clients.
| Tool | Auth | Description |
|---|---|---|
list_supported_tokens | None | Query supported tokens and networks |
get_payment_quote | None | Fee estimates for any payment |
create_payment | JWT | Create a payment proposal |
check_payment_status | JWT | Check payment status by ID |
list_payments | JWT | List recent payments |
create_invoice | JWT | Generate invoice with payment link |
list_invoices | JWT | List invoices |
get_balance | JWT | Query wallet balances per network |
Two transport modes: Streamable HTTP (POST /api/mcp) for web-based clients and stdio (pnpm mcp:stdio) for Claude Desktop and CLI integrations. Public tools (token list, quotes) require no authentication; all payment operations require a valid JWT.
Ready to build AI-native payments?
Protocol Banks is open-source and ready for integration. Start with the AI Wallet SDK, connect your agent via MCP, or explore the A2A protocol for autonomous agent collaboration.