AI agents and online stores: what to do now and what can wait

AI agents already search and compare products for shoppers, and the space now has so many protocols and promises that it is hard for a store owner to know what to do. In short: most stores do not need to rush into a protocol. But there is work that pays off whatever happens, and it is worth starting now.

Three protocols in plain words

• MCP connects an AI agent to tools and data. It is not a buying protocol in itself, but a connection layer.
• ACP comes from OpenAI and Stripe and focuses on checkout and payment.
• UCP is the open standard Google and Shopify presented in January 2026. It covers cart, checkout, account linking and order tracking. A store publishes a profile at /.well-known/ucp saying what it supports. It can connect over REST, MCP, A2A or an embedded mode. The store usually remains the merchant of record.

The protocols do not cancel each other out: UCP has an MCP binding, so they are closer to layers than rivals.

What practice has shown

Buying right inside a chat turned out to be harder than promised. In March 2026 OpenAI scaled back Instant Checkout: a spokesperson said it was moving to apps inside ChatGPT. ACP carries on. The takeaway for a store: AI search and recommendations matter today, while purchases inside a chat are still an experiment.

Where the platforms stand

• Shopify is ahead. In May 2026 the admin got a dedicated Agentic Storefronts page, and products reach AI channels through Shopify Catalog. Which channels are on for you depends on factors such as country, so check your admin.
• Adobe Commerce committed to UCP and ACP in February and showed a Commerce MCP server at Summit. As of August 2026, partners reported no native UCP support.
• Magento Open Source: we found no official solution. Connecting it to agents will most likely be custom work on top of the GraphQL or REST API.

What to do now

1. Clean up product data. Complete attributes, clear descriptions, tidy categories, barcodes and SKUs. An agent compares data, not banners. This is the best-value preparation on any platform.
2. Keep prices and stock accurate. A person may notice an error on the storefront, but an agent will pass it to the shopper as fact.
3. Check your APIs. Look at GraphQL and REST speed and stability, and at caching: Varnish and CDNs must not get in the way of future agent endpoints.
4. Review robots.txt and bot protection. Do not block everything, or you may lose visibility. Do not open everything either: decide which bots can reach what.
5. Set up analytics. Put traffic from AI services in its own segment, keeping in mind that not every agent identifies itself.
6. Think about security. A product description or review can hide an instruction for an agent. Agents need minimal permissions and limits, and risky actions need confirmation.

When to start an integration

On Shopify, start with the platform settings. On Magento and custom platforms, an integration makes sense if you have a large catalog, international sales or B2B repeat orders. Start small: a read-only MCP server for catalog and availability, with results measured. Checkout can come later, once the protocols and your platform settle.

We have worked with Magento for more than ten years, and we build Shopify stores and MCP servers: we will review your store's data and APIs and tell you what is worth doing now and what can wait.

Sources: UCP documentation, Stripe on UCP, Digital Commerce 360 on Instant Checkout, Shopify changelog, Inchoo on the state of agentic commerce, Kinex Media on UCP in Adobe Commerce.