For a Shopware merchant, agentic commerce begins with a decision: should agents help customers discover the catalog, carry out a purchase or take on work inside the business? Each goal calls for a different approach. Understanding the distinction makes it easier to start with a manageable project and build on it later.
This guide connects the Shopware documentation with a proposed approach to implementation. A suitcase that a customer needs before a trip serves as a worked example throughout. What matters is the right variant, the delivery date and an order that reaches the shop correctly.
Choose the starting point around a customer task
| Your first goal | Where to start | Outcome of the first trial |
|---|---|---|
| Customers should discover suitable products through an external assistant | Product data and a product feed | The assistant finds the correct variant with suitable information |
| An agent should handle a purchase in the shop | UCP access for the selected sales channel | An authorised test purchase produces the correct order |
| An agent should take on work in sales or operations | A focused experiment with a suitable Lab project | The team can delegate a specific task and check the result |
These approaches can work together. For the first implementation, connect one of them to a clear task. “Customers should find our suitcases” requires different preparation from “An agent should order the selected suitcase.”
Configure the product feed and UCP separately
The Shopware documentation for the Agentic Commerce extension describes a product feed and UCP from extension version 1.1.0. It lists the extension as beta. The feed has its own sales channel; UCP is configured in the Agentic Commerce tab of a Storefront or Headless sales channel. The UCP profile is published at /.well-known/ucp, where agents discover the capabilities offered. Documented transports include REST, A2A, Embedded and, from Shopware 6.7, MCP. The documentation currently lists the US as the available market for ChatGPT Marketplace registration.
For planning, decide which assistant and market you want to reach before configuring the connection. Then work with your implementation partner to check which approach supports the intended task. The linked documentation provides the controls and requirements for the version you use.
A feed helps customers discover an offer. During the purchase, the current price, selected variant and delivery option need to agree again. The UCP specification describes capability negotiation and workflows between the participating systems. The protocol comparison explains how UCP, MCP and ACP fit into the purchasing journey.
Start with a small catalog that reveals real decisions
Choose products that expose the decisions your customers regularly make. For a suitcase, these include dimensions, weight, available variants and the desired delivery date. In spare parts, they might include the connector, model year and approved alternatives. The B2B guide adds customer prices, quotations and approvals.
Record which product attributes change the selection in your first trial. An agent should rule out an oversized suitcase as reliably as one that would arrive after the trip. Check the feed, product page and order flow together. The product data freshness guide helps with prices, stock and conflicting information.
Follow the test through the order and customer service
Decide beforehand what a successful task looks like. Adapt the following template to your own shop. The following steps are recommended for your first test.
| Test in your own shop | What to check afterwards |
|---|---|
| The customer chooses a suitcase of a particular size | The proposal and the order contain the same suitable variant |
| The price changes before confirmation | The customer sees the valid total and approves the changed order |
| The desired variant is no longer available | The flow leads to a new selection or a clear handover |
| Confirmation is repeated after a connection failure | The confirmed task produces exactly one order |
| The customer asks about delivery | Customer service can identify the order and its current status |
Check the order in the Shopware Administration and follow what the connected systems do with it. An agent can tell a customer that a purchase is complete; the merchant needs the order, payment and shipping to fit together afterwards. The eight checks for agent purchases provide further cases.
Divide the work between merchant, partner and Lab
The merchant knows the customer questions and decides which decisions may be delegated. The implementation partner connects data, the shop and existing workflows. They also test the exceptions that consume time in daily operations: customer terms, incomplete addresses, changed deliveries or a return. This is where good implementation creates value.
The Shopware Agentic Commerce Lab provides projects for exploring new approaches. Choose a project around the task you want to understand. A price negotiation calls for a different experiment from product selection. Bring the findings into your own architecture and acceptance process.
Make the next decision before expanding
After the first trials, collect what helped or interrupted the customer's task. Was any information unclear? Did the customer need to repeat too much? How much work remained with sales and customer service? Use the findings to decide whether better data, another purchasing route or greater autonomy for the agent would help next.
For a broader starting point, read the guide to preparing for agentic commerce. The free handbook connects implementation with strategy, experience and the move towards autonomous shop operations.