Shop MCP: Catalog Search and Shop Pay Checkout for Agents

Shop MCP is a Shopify MCP server that lets an agent search Shopify Catalog, build carts and complete Shop Pay checkout through one connection. Shopify announced it in its developer changelog on 13 October 2026. It works with over 1 billion products in Shopify Catalog, and search needs no API key.

Nothing changes for existing Catalog API integrations, and no action is required. The change matters to anyone who builds or configures an MCP-compatible agent that should be able to buy, not just browse.

What changed with Shop MCP?

Before Shop MCP, an agent could find a product in Catalog but had to send the shopper to the store's website to pay, which Shopify says often caused drop-off. Now a signed-in shopper can pay with the Shop Pay details they have already saved, without leaving the agent.

The server is live at https://mcp.shop.com, with a skill description that MCP-compatible agents can read. One connection gives an agent 17 tools in five main flows:

• Search and product details: search_catalog, lookup_catalog and get_product read from Shopify Catalog.
• Sign-in with Shop: the shopper signs in once, and search tools use their saved sizes and preferences. get_account confirms the connection.
• Carts and checkout on Shopify stores: the agent builds a cart and creates a checkout at the store that sells the product, with create_cart, get_cart, update_cart, create_checkout, get_checkout, update_checkout and complete_checkout. Both work without sign-in.
• Checkout on other stores: when a store does not run on Shopify, the shopper approves the purchase in Shop and the agent receives single-use card details to pay on that store's checkout. Shop's skill file runs this through five spend-request tools, from create_spend_request to complete_spend_request.
• Orders: for a signed-in shopper, the agent finds orders and tracking with search_orders.

Orders placed through an agent appear in the merchant's Shopify admin like any other order. For a merchant, that means agent orders go through the normal fulfilment flow. Check how they are attributed in reports before relying on channel figures.

Shop MCP also means the catalogue data a merchant already keeps is what an agent sees. A search result shows the store, the price and whether the item is in stock, so stale stock levels or prices become wrong answers inside someone else's agent. Keeping inventory and pricing in sync matters more once agents can buy on the shopper's behalf.

How a Shop Pay checkout runs through the agent

A signed-in shopper is what turns a cart into a Shop Pay purchase. With one, checkout can use their saved addresses and Shop Pay, and the agent completes the purchase with complete_checkout after the shopper confirms the total and payment method. MCP clients such as Claude and ChatGPT run the sign-in for you, and your own code can use the device flow.

Sign-in is optional. Without it, Shop's skill file tells the agent to return each merchant's complete continue_url alongside its items, so the shopper finishes on the store's own checkout. There are no webhooks at launch, so the agent has to poll get_checkout for checkout status. Shopify's test for a working purchase is a checkout that returns status: completed with an order. The skill file adds a rule worth copying into your own code: never repeat complete_checkout, and settle an uncertain completion only by reading get_checkout.

Plan the polling loop early, before you polish the happy path. It decides how the agent reports progress to the shopper, and it is easy to leave out when purchases are only tested by hand.

A worked example: a gift-finder agent

Take a gift-finder agent that a shopper asks for a ceramic coffee dripper. Walk it through the flows in order.

1. Search. The agent calls search_catalog. Each result shows the store, the price and whether the item is in stock.
2. Sign-in. The shopper signs in with Shop, so search uses their saved preferences and checkout can use their saved address and Shop Pay.
3. Cart and checkout. The agent builds the cart and creates the checkout at the store that sells the dripper.
4. Confirm and pay. The shopper confirms the total and payment method, and the agent calls complete_checkout.
5. Poll. The agent polls get_checkout until it returns status: completed with an order.
6. Track. Later, the agent answers "where is my order?" with search_orders.

If the shopper wants to buy from a store that does not run on Shopify, steps 3 to 5 change. The shopper approves the purchase in Shop, and the agent pays on that store's own checkout with the single-use card details it receives. In the skill file, the agent reports the result with complete_spend_request, as a status of success or error, instead of polling get_checkout. Our guide to agentic checkout architecture covers scoped payment tokens that keep card numbers off your servers, which is worth comparing with this card-details path.

What a live search returned

We called the server directly on 14 October 2026. Asking it to list its tools returned exactly the 17 names above. A search_catalog call for "ceramic coffee dripper", limited to three results, came back in UCP format, tagged with UCP version 2026-08-25.

Each product carried a rating, media and its variants. Each variant named its seller, its price, its availability and a checkout_url with payment=shop_pay on the seller's own domain. Prices arrived in minor units: 3500 in USD is $35.00. The three results came from three different US stores, all priced in USD. We sent no buyer location with the request.

Two details matter when you build against it. The server only accepted MCP protocol version 2026-07-28, and rejected an older version with "Unsupported protocol version", so check your MCP client library first. And test with your own shoppers' country and currency before assuming the results will suit UK buyers.

Shop MCP or Shopify's Catalog MCP servers?

Shop MCP is not the only Shopify MCP endpoint, and the others work differently. Pick by what the agent has to do, rather than by which one sounds newest.

Global Catalog MCP implements the Universal Commerce Protocol (UCP) Catalog capability for discovery across multiple Shopify merchants, such as comparison shopping. Shopify points single-store agents to Storefront Catalog MCP instead. Both catalogue endpoints require an agent profile URL in meta.ucp-agent.profile on every request, while Shop MCP search needs no credentials. Our Shopify MCP server guide covers the catalogue servers in depth, and our explainer on how ACP, MCP and UCP fit together covers the protocols underneath.

What should you do first?

Start in the terminal, not in code.

1. Try a search. Install the CLI with npm install -g @shopify/shop, then run shop search with a product your shoppers would look for to see the data shape.
2. Connect the agent. Point your coding agent at the skill file, or add https://mcp.shop.com as an MCP server in its config.
3. Add Shop sign-in. Let the MCP client handle it, or use the device flow in your own code.
4. Test one purchase end to end, and only call it working when get_checkout returns status: completed with an order. This places a real order at a real store, so buy something cheap you are willing to keep or return.

If you want an agent like this scoped and built against your own catalogue, our Shopify integrations team can help.