# Agentic Commerce

Enable AI agents to make autonomous payments. The future of commerce is here.

## Overview

Agentic Commerce enables AI agents to make payments and conduct transactions autonomously on behalf of users. Built on the Model Context Protocol (MCP) and secure payment infrastructure, it opens up entirely new possibilities for AI-driven commerce.

## What is Agentic Commerce?

Agentic Commerce is a payment framework that allows AI agents to:

- **Make purchases** on behalf of users (with authorization)
- **Process payments** for services rendered
- **Manage subscriptions** and recurring billing
- **Handle refunds** and dispute resolution
- **Track spending** and provide transparency

### Key Principles

1. **User Authorization**: Every agent operates within user-defined limits
2. **Transparency**: All transactions are logged and visible
3. **Control**: Users can revoke access at any time
4. **Security**: Bank-level authentication and encryption

## Use Cases

### Personal AI Assistants

Your AI assistant can:
- Book travel (flights, hotels, car rentals)
- Order groceries and household items
- Pay bills and manage subscriptions
- Make restaurant reservations with deposits
- Purchase event tickets

**Example**:
```
User: "Plan a weekend trip to Vienna for next month"
Agent: Books flights, hotel, and restaurant reservations
       Processes payments within approved budget
       Sends confirmation and itinerary
```

### Business Process Automation

AI agents can:
- Procure office supplies automatically
- Pay vendor invoices
- Manage SaaS subscriptions
- Handle employee expense reimbursements
- Process customer refunds

**Example**:
```
Agent monitors inventory levels
Automatically reorders when stock is low
Processes payment via approved vendor
Updates accounting system
```

### Customer Service Bots

Service agents can:
- Process refunds instantly
- Issue store credits
- Upgrade subscriptions
- Apply discounts and promotions
- Handle warranty claims

**Example**:
```
Customer: "My order arrived damaged"
Agent: Verifies claim via photo
       Issues instant refund or replacement
       Processes payment reversal
```

### Research & Data Services

AI agents can:
- Purchase research reports
- Access paid APIs and datasets
- Subscribe to premium content
- Commission custom analysis
- Pay for compute resources

## How It Works

### Architecture

```
┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│    User     │────▶│  AI Agent    │────▶│  BANKpay+   │
│  (Principal)│     │  (MCP)       │     │  Payment    │
└─────────────┘     └──────────────┘     └─────────────┘
       ▲                    │                    │
       │                    ▼                    │
       │            ┌──────────────┐            │
       └────────────│  MCP Server  │◀───────────┘
                    │  (Registry)  │
                    └──────────────┘
```

### Flow

1. **User Authorization**
   - User grants spending authority to agent
   - Sets limits (amount, category, time period)
   - Defines allowed merchants/categories

2. **Agent Action**
   - Agent identifies need for payment
   - Checks against authorization limits
   - Initiates payment via MCP protocol

3. **Payment Processing**
   - BANKpay+ validates agent credentials
   - Verifies user authorization
   - Processes payment instantly

4. **Confirmation & Logging**
   - Transaction recorded in user's ledger
   - User receives notification
   - Agent receives confirmation

## MCP Integration

### Model Context Protocol

Agentic Commerce is built on the **Model Context Protocol (MCP)**, an open standard for AI agent interactions.

#### MCP Payment Extension

```typescript
// Agent requests payment capability
{
  "protocol": "mcp",
  "extension": "payments",
  "version": "1.0",
  "capabilities": [
    "initiate_payment",
    "check_status",
    "process_refund"
  ],
  "authorization": {
    "user_id": "user_123",
    "agent_id": "agent_456",
    "limits": {
      "max_amount": 500,
      "currency": "EUR",
      "categories": ["travel", "supplies"],
      "valid_until": "2026-12-31"
    }
  }
}
```

### Getting Started with MCP

1. **Register your agent** on the MCP server
2. **Implement payment capabilities** using our SDK
3. **Request user authorization** via OAuth flow
4. **Start processing payments** within approved limits

[View MCP Documentation](/developers/docs/agentic)

## Security & Controls

### User Controls

| Control | Description |
|---------|-------------|
| **Spending Limits** | Set maximum amount per transaction/day/month |
| **Category Restrictions** | Limit to specific merchant categories |
| **Merchant Allowlist** | Approve specific merchants only |
| **Time Limits** | Authorization expires after set period |
| **Transaction Notifications** | Real-time alerts for all payments |
| **Instant Revocation** | Revoke access immediately |

### Technical Security

- **OAuth 2.0**: Secure authorization flow
- **Signed Requests**: All agent requests are cryptographically signed
- **Audit Trail**: Complete transaction history
- **Fraud Detection**: AI-powered anomaly detection
- **SCA Compliance**: Strong Customer Authentication

### Agent Verification

All agents must be:
- **Registered**: Unique agent ID on MCP server
- **Verified**: Identity verification for agent operators
- **Bonded**: Security deposit for high-volume agents
- **Insured**: Liability coverage for errors

## For Developers

### API Example

```python
from bankpay import AgenticCommerce

# Initialize client
client = AgenticCommerce(api_key='your_key')

# Register agent capabilities
agent = client.register_agent(
    agent_id='travel_assistant_v1',
    capabilities=['payment', 'refund'],
    max_amount=1000,
    categories=['travel', 'hospitality']
)

# Process payment
payment = client.initiate_payment(
    agent_id='travel_assistant_v1',
    user_id='user_123',
    amount=450.00,
    currency='EUR',
    merchant='Austrian Airlines',
    category='travel',
    description='Flight booking VIE-MUC'
)

print(f"Payment status: {payment.status}")
```

### SDKs Available

- **Python**: `pip install bankpay-agentic`
- **TypeScript**: `npm install @bankpay/agentic-commerce`
- **Go**: `go get github.com/bankpay/agentic`
- **Rust**: `cargo add bankpay-agentic`

### Webhooks

Real-time notifications for:
- `agent.registered`
- `authorization.granted`
- `authorization.revoked`
- `payment.initiated`
- `payment.completed`
- `payment.failed`
- `limit.reached`
- `fraud.detected`

## Pricing

Agentic Commerce pricing is based on transaction volume.

| Plan | Transactions/Month | Price | Features |
|------|-------------------|-------|----------|
| **Developer** | Up to 100 | Free | Sandbox access, basic docs |
| **Startup** | Up to 1,000 | €99/month | Production access, email support |
| **Growth** | Up to 10,000 | €499/month | Priority support, custom limits |
| **Enterprise** | Unlimited | Custom | Dedicated support, SLA, custom features |

**Transaction Fees**:
- Standard: 0.25% per transaction
- Volume discounts available
- No setup fees

[Contact sales](/contact) for enterprise pricing.

## Compliance & Regulation

### Regulatory Framework

Agentic Commerce operates within existing payment regulations:

- **PSD2**: Compliant with Payment Services Directive 2
- **GDPR**: Full data protection compliance
- **AML**: Anti-money laundering procedures
- **KYC**: Know Your Customer verification

### Agent Liability

Clear liability framework:
- **User**: Responsible for authorized transactions
- **Agent Operator**: Liable for unauthorized actions
- **BANKpay+**: Payment processing only, not agent actions

### Dispute Resolution

- **Automated**: AI handles common disputes
- **Mediation**: Human review for complex cases
- **Insurance**: Coverage for agent errors
- **Legal**: Clear terms of service

## Real-World Examples

### Example 1: Travel Booking

```
User: "Book me a business trip to Berlin"

Agent Actions:
1. Searches flights (Lufthansa, €350)
2. Books hotel (3 nights, €450)
3. Rents car (3 days, €180)
4. Total: €980 (within €1000 limit)

Payments:
- Flight: ✅ Charged €350
- Hotel: ✅ Charged €450
- Car: ✅ Charged €180

User receives:
- Itinerary with all confirmations
- Itemized expense report
- Receipt copies
```

### Example 2: Procurement

```
System detects low inventory:
- Item: Office chairs (5 units)
- Approved vendor: Office Depot
- Budget: €1000

Agent Actions:
1. Creates purchase order
2. Processes payment (€875)
3. Schedules delivery
4. Updates inventory system

Result:
- Inventory replenished
- Payment processed
- Accounting updated
```

### Example 3: Customer Service

```
Customer: "I want to return this item"

Agent Actions:
1. Verifies purchase (order #12345)
2. Confirms return eligibility
3. Processes refund (€129.99)
4. Sends return label

Result:
- Customer satisfied
- Refund processed instantly
- Return initiated
```

## Getting Started

### For Agent Developers

1. **Sign up** at [bankpay.plus/agentic-commerce](/agentic-commerce)
2. **Register your agent** on the MCP server
3. **Implement payment flow** using our SDK
4. **Test in sandbox** with simulated payments
5. **Go live** and start processing real payments

### For Users

1. **Enable agentic payments** in your dashboard
2. **Authorize agents** you want to use
3. **Set spending limits** and restrictions
4. **Monitor transactions** in real-time
5. **Revoke access** anytime

## Resources

### Documentation

- [Agentic Commerce Overview](/developers/docs/agentic)
- [MCP Integration Guide](/developers/mcp)
- [API Reference](/developers/api/agentic)
- [Security Best Practices](/developers/security/agentic)
- [Sample Implementations](/developers/examples)

### Tools

- **MCP Server**: [mcp.bankpay.plus](https://mcp.bankpay.plus)
- **Agent Registry**: Browse registered agents
- **Sandbox**: Test environment for developers
- **Webhook Tester**: Debug your integrations

### Community

- **Discord**: Join developer community
- **GitHub**: Open-source SDKs and examples
- **Forum**: Discuss best practices
- **Blog**: Latest updates and case studies

## FAQ

### Is this safe?

Yes. Users maintain full control with spending limits, category restrictions, and instant revocation. All transactions are logged and visible.

### What happens if an agent makes a mistake?

Agent operators are liable for unauthorized transactions. We also offer insurance coverage for agent errors.

### Can I use this with any AI?

Any AI agent that implements the MCP protocol can use Agentic Commerce. We provide SDKs for popular frameworks.

### How do I track what my agent is doing?

Real-time notifications for all transactions. Full audit trail in your dashboard. Export to accounting software.

### Can I set different limits for different agents?

Yes. Each agent authorization is independent with custom limits, categories, and time periods.

### What if I want to stop an agent?

Instant revocation. The agent can no longer initiate payments. Existing authorizations are voided immediately.

## The Future of Commerce

Agentic Commerce represents a fundamental shift in how payments work:

- **From manual to automatic**: No more clicking "buy"
- **From reactive to proactive**: Agents anticipate needs
- **From fragmented to unified**: One authorization model
- **From opaque to transparent**: Complete visibility

We're building the infrastructure for an AI-driven economy.

## Ready to Enable Agentic Payments?

[**Get Started**](/agentic-commerce) - Free developer account

Or [**Contact Sales**](/contact) for enterprise solutions.

---

**Last updated**: February 28, 2026

**Learn more**: [Agentic Commerce PR](/agentic-commerce-pr)

**Questions?** Contact us at agentic@bankpay.plus