AP2 Human-Present Transaction Flow

Before writing code

Fetch live docs:

1. Fetch https://ap2-protocol.org/ap2/specification/ for the human-present flow specification
2. Web-search site:github.com google-agentic-commerce AP2 samples human-present for reference implementations
3. Fetch https://github.com/google-agentic-commerce/AP2/tree/main/code/samples for the card payment sample
4. Fetch https://ap2-protocol.org/overview/ for flow overview

Conceptual Architecture

What Human-Present Means

In a human-present flow, the user is actively engaged throughout the transaction. They review products, select items, choose a payment method, and confirm the purchase — all while interacting with the Shopping Agent.

The Human-Present Transaction Flow

The official AP2 specification describes approximately 11 high-level steps for the human-present flow. The primary VDC in this flow is the Checkout Mandate.

Phase 1: Shopping Intent
  1. User → Shopping Agent:    Provides shopping prompt ("I want to buy a coffee maker")
  2. SA → User:                Confirms intent and collects CP preference + shipping address

Phase 2: Product Discovery
  3. SA → CP:                  Queries Credentials Provider for available payment methods
  4. SA → Merchant:            Presents shopping intent
  5. Merchant → SA:            Creates and signs Checkout Mandate(s) with product offers

Phase 3: User Confirmation
  6. SA → User:                Displays final cart + payment options
  7. User → SA:                Reviews cart, selects payment method, confirms on trusted device surface

Phase 4: Payment Processing
  8. SA → Merchant:            Sends confirmed Checkout Mandate + user attestation
  9. Merchant → MPP:           Submits payment for processing
  10. MPP:                     Constructs Payment Mandate and requests credentials from CP

Phase 5: Completion
  11. MPP → Merchant → SA → User: Payment processed, receipt delivered

Agent Interactions (A2A Messages)

Each step involves A2A protocol communication:

• Shopping Agent ↔ Merchant Agent: A2A tasks with mandate DataParts
• Shopping Agent ↔ Credentials Provider: A2A tasks for payment methods
• Shopping Agent ↔ Payment Processor: Indirect (via Merchant)

User Touchpoints

The user is involved at these critical points:

• Step 1: Initiates the shopping request
• Step 2: Confirms intent and provides preferences
• Step 6-7: Reviews cart, selects payment, and confirms on trusted device surface (attestation)
• Step 11: Receives confirmation and receipt

Trusted Device Surface

Step 7 is a load-bearing security step:

• User is redirected to a trusted device surface (not the agent)
• User reviews the final transaction details
• User provides attestation (biometric, PIN, etc.)
• This is hardware-backed confirmation, not just an LLM prompt
• Prevents agent manipulation of the final authorization

Challenge Handling

During the flow, any participant may trigger a challenge:

• 3DS2 — Card network challenges for risk assessment
• OTP — One-time password verification
• User is redirected to a trusted surface for challenge resolution
• Flow resumes after successful challenge

Implementation Considerations

• Stateful flow — The human-present transaction flow must maintain state across multiple agent interactions
• Timeout handling — Users may step away; implement reasonable timeouts
• Error recovery — Handle failures at any step with proper rollback
• Concurrent agents — Shopping Agent may query multiple merchants in parallel
• Cart changes — User may modify the cart between steps

Best Practices

• Display clear product information at step 6 for user review
• Never skip the trusted device confirmation (step 7)
• Handle 3DS/OTP challenges gracefully
• Log every step for audit trail
• Implement timeouts for user response steps
• Support cart modification before final confirmation
• Show the user exactly what they're authorizing (items, prices, total)
• Test the full human-present transaction flow end-to-end

Fetch the specification and sample implementations for exact message formats, mandate structures at each step, and agent communication patterns before implementing.