Chapter 11 10 min read

The End-to-End Agentic Workflow

A step-by-step walkthrough of the purchase of tomorrow.

We have discussed the components—MCP for data, ACP for payments, Shopify for supply. Now, let's watch them all work in concert.

In this scenario, a user named Alex is chatting with a general-purpose AI assistant (like ChatGPT or Claude) and wants to buy a new backpack.

Phase 1: Intent & Discovery (User → Agent → MCP)

User (Alex):

"I need a waterproof backpack for commuting. Something stylish, under $150. Check what's available on independent stores."

The Agent's Thought Process:

  1. Analyze intent: Product Search (Backpack). Constraints: Waterproof, Stylish, <$150. Source: Independent (Shopify).
  2. Action: Call Shopify_MCP_Server.search_products().
// MCP Request { "query": "waterproof commuter backpack", "price_max": 150, "limit": 3 }

The Shopify MCP Server receives this. It doesn't return HTML. It returns structured JSON of 3 candidate products from different stores, including real-time stock status.

Phase 2: Recommendation & Reasoning (Agent → User)

The Agent doesn't just dump the JSON. It uses its reasoning capabilities to curate the options.

Agent:

"I found three great options for you:"

1. Rains Backpack Mini $95.00

Best for pure waterproofing. Minimalist style.

2. Bellroy Classic $139.00

Best organization. Water-resistant.

3. Topo Designs Daypack $119.00

Rugged look. Good for weekends too.

"I recommend the Bellroy because you mentioned commuting, and it has a dedicated laptop sleeve."

Phase 3: Selection & Action (User → Agent)

User (Alex):

"Good call. Let's go with the Bellroy. Use my main card."

The Agent now switches from Explorer Mode (MCP) to Transaction Mode (ACP).

Phase 4: Execution & Payment (Agent → ACP → Stripe)

This is the critical step. The user does not leave the chat. They do not go to bellroy.com.

The Agent's Action:

  1. Retrieve user's shipping info (from memory/context).
  2. Request a payment token. Since the user said "main card," the Agent asks the Wallet Provider (Stripe/PayPal) for a one-time virtual card or token authorized for $139.00.
  3. Construct an ACP Order Message.
// ACP Order Payload { "intent": "PURCHASE", "product_id": "bellroy_classic_v2_black", "merchant_id": "bellroy_shopify_store", "shipping_address": { ... }, "payment_token": "tok_visa_virtual_12345", "amount": 139.00 }

The Agent sends this payload to the merchant's ACP endpoint.

The Merchant's System:
1. Validates the token.
2. Processes the charge.
3. Creates the order in Shopify.
4. Returns a success message.

Phase 5: Confirmation (Agent → User)

Agent:

Order Confirmed!

"I've ordered the Bellroy Classic. It will arrive at your home address by Thursday. Order #SH-998877."

Total Time: 45 seconds.
Clicks: 0.
Apps Opened: 1 (The Chat).

Key Takeaways

  • The workflow unifies Discovery, Recommendation, and Transaction into a single conversational stream.
  • MCP is used to fetch live product data and inventory.
  • ACP is used to execute the payment securely without exposing the user's real card.
  • The user remains in control of the decision ('Go with the Bellroy'), but delegates the execution.
Next: Optimizing Product Data for Agents