Imagine a customer opening ChatGPT and asking: “Find me a blue backpack from your shop for less than €100, add the 20-litre version to my cart, and help me check out.”
With the right integration, the assistant can search your store's current inventory, show matching products, check available options, build a real cart, and guide the customer through checkout. The customer shops through a conversation, while your ecommerce system continues to manage the products, prices, stock, and orders.
An MCP server makes your shop's features available to the assistant. It connects the shopping conversation to your store's APIs. Through a supported ChatGPT plugin or Claude connector, customers can use those features without navigating each step of your website themselves.
Here is how the complete experience works, from the first product search to payment and order tracking.
What does it mean to make your shop available in ChatGPT?
It means giving the assistant the ability to interact with your store, beyond answering questions about your brand. A customer can ask what is in stock, choose a specific product, change their cart, and start checkout. Each action calls a function connected to your ecommerce backend.
The MCP server provides those functions as tools. Your connector or plugin makes them available in the assistant. An optional interface can display product images, colour choices, and cart summaries directly in supported chat surfaces.
Your shop still controls the transaction. Its systems decide the current price, whether a variant is available, which discounts apply, and whether an order has been paid.
The journey below is an illustrative design for a physical-goods shop. Platform details were checked on October 6, 2026.
1. Search your store's inventory in conversation
Customer: “Show me waterproof backpacks under €100 that are available in blue.”
The assistant calls your product-search tool with the customer's requirements. Your MCP server queries the shop's catalogue and inventory services, then returns matching products with their names, images, prices, available variants, and product links.
This is a search of your store's data. If the blue version is sold out, the response should reflect that. If a bag is described as water-resistant rather than waterproof, the assistant should preserve that distinction.
The customer can refine the search naturally:
“Only show me the ones that fit a 15-inch laptop.”
“Do you have something lighter?”
“What is the difference between these two?”
To support those questions, expose the relevant attributes in your product API: dimensions, materials, capacity, compatibility, and other details shoppers actually use to decide. The assistant can compare what your catalogue provides; it should not invent missing specifications.
2. Choose a product, size, colour, and quantity
Customer: “I like the second one. Do you have it in blue, in the 20-litre size?”
The assistant retrieves the product's variants and checks the exact combination. A blue 20-litre backpack and a black 30-litre backpack may belong to the same product page, but they are different purchasable items.
Your MCP server returns the correct variant identifier, price, and availability. If the requested combination does not exist, the assistant can show the alternatives your store actually offers.
This is also where customers can ask buying questions: whether a laptop will fit, what the warranty covers, or how returns work. Product-detail and shop-policy tools let the assistant answer from your maintained information.
The result is a specific selection the customer understands: one blue, 20-litre backpack at the current quoted price.
3. Add products to a real shopping cart
Customer: “Add one to my cart, and include the matching rain cover.”
The assistant calls a cart tool. The MCP server asks your ecommerce system to create or update a cart using the selected variant IDs and quantities. It returns the actual cart state, including line items, discounts, and the amounts your backend has calculated.
The customer can continue shopping in the same conversation:
“Make that two backpacks.”
“Remove the rain cover.”
“Apply my discount code.”
“What is in my cart now?”
Each request maps to a supported store operation. The assistant does not just write a shopping list in the chat: it updates the cart that will be used at checkout.
Keep the cart associated with the correct shopper or guest session. Reuse it across the shopping journey, and return the updated contents after every change. If a request is retried after a timeout, your integration should avoid accidentally adding the same item twice. See our guide to idempotency for agent actions.
4. Calculate delivery options and review the total
Customer: “Can you deliver to Lyon? What would the total be with express shipping?”
If your commerce API supports delivery quotes, the MCP server can expose that feature too. The assistant requests the information needed for a quote, retrieves the available delivery options, and presents the resulting costs and estimates.
Your backend calculates discounts, shipping, and applicable taxes. The assistant explains those results. When an address or another required detail is missing, it should identify what remains to be calculated.
Before checkout, show a clear summary: the selected variants, quantities, delivery method, currency, and final amount where available. Recheck availability and prices through the commerce system so that an earlier product recommendation is not treated as a permanent quote.
5. Check out and complete the purchase
Customer: “Everything looks right. Let's check out.”
Your MCP server creates or retrieves a checkout session for the cart. There are two experience patterns to consider.
Continue on your shop's checkout. The assistant provides a link to the customer's prepared checkout. The shopper opens it, reviews the order, enters payment information, and completes the purchase using your existing checkout system. Product discovery and cart building happened in chat; payment finishes on your store.
Complete checkout within a supported chat interface. Where the platform and payment integration support it, a shopper can review and confirm the purchase inside the conversation's interface. This requires the relevant checkout integration in addition to exposing an MCP tool.
OpenAI currently recommends external checkout for eligible physical-goods plugins. Its documentation also describes checkout using payment methods already saved with the merchant, and a ChatGPT payment sheet available to select partners in private beta. The saved-method option cannot collect new payment credentials. See the OpenAI checkout documentation.
The distinction matters when you describe your customer experience: an MCP server can expose checkout operations, but connecting it alone does not activate every host's payment interface. Use the supported flow for each platform, and keep payment details within that checkout system.
Once the merchant confirms the order, the assistant can return the order reference and next steps. Creating a checkout session is not the same as completing a purchase.
6. Track the order and support the customer after purchase
The experience can continue after checkout:
“Has my order shipped?”
“Where is my parcel?”
“Can I return the rain cover?”
If your APIs support these operations, you can expose order status, tracking, return eligibility, return requests, and customer support through MCP as well.
Account-specific tools need the customer's authenticated identity and permission checks. Your server must verify that an order belongs to the caller before returning its details or changing it. A customer should never gain access to another person's purchase simply by providing an order number.
Which ecommerce features can you expose through MCP?
MCP is not limited to product search. You can expose the full set of store features supported by your APIs and appropriate for the customer and host platform. Each capability needs a tool implementation, permissions, and a clear result—not just a name in a tool list.
- Catalogue and inventory search: “Which running shoes are in stock in my size?”
- Product details and comparisons: “Compare these two models.”
- Variants and availability: “Is the green version available in medium?”
- Cart creation and updates: “Add two, then remove the other item.”
- Discounts and loyalty benefits: “Can I use my loyalty points on this order?”
- Shipping and tax estimates: “What is the delivered price to my address?”
- Checkout: “Prepare my order so I can pay.”
- Order history and tracking: “Show me the status of my last order.”
- Returns and support: “Start a return for the item that arrived yesterday.”
These are possible integrations, not features every platform provides automatically. Expose shopper capabilities to shoppers; keep stock administration, pricing changes, and other staff functions behind separate staff permissions.
How to connect your ecommerce shop to ChatGPT and Claude
Your developer starts by connecting the MCP server to the commerce APIs behind your store. They define tools for the shopping journey, map each tool to the corresponding API operation, and deploy a remote HTTPS endpoint. Our OpenAPI-to-MCP guide explains how existing API documentation can help with that work.
For Shopify-based integrations, check the platform's commerce interfaces before recreating them. Shopify documents cart and checkout MCP operations, including a handoff to merchant-hosted checkout. See its checkout integration guide.
Make the shop available in ChatGPT
OpenAI's current model packages MCP connections with optional skills and interfaces as plugins. Its quickstart explains how to add a custom MCP server, create and install a personal plugin, and test it in ChatGPT Work. Use that flow to validate your shop with test products and orders. See the ChatGPT plugin quickstart.
For public distribution, package the integration and follow the submission and review process. Customers need access to the supported plugin experience; simply hosting an MCP server does not automatically make your shop available to every ChatGPT user or place it in shopping recommendations.
Make the shop available in Claude
Claude supports custom connectors backed by remote MCP servers. Customers connect the server through the connector settings and complete the authentication your shop requires. Organization-managed accounts may need an administrator to add the connector first. Follow Claude's custom connector guide.
You can reuse your commerce backend and much of the MCP implementation across both platforms. Test the complete journey in each: product search, variant selection, cart changes, checkout, and order confirmation. Optional interfaces and payment experiences may differ.
Turn your existing shop into a conversational shopping experience
A useful ecommerce MCP integration carries the customer through a complete purchase journey. They describe what they want, search your inventory, choose the right item, add it to a real cart, review delivery and pricing, and complete a supported checkout flow. Order tracking and returns can follow through the same connected service.
Start by listing the features your customers already use on your website and checking which are available through your APIs. That becomes the scope of your MCP integration—and the shopping experience you can bring to ChatGPT and Claude.
The Agent Readiness Audit starts from your public API documentation and proposes a tool scope, readiness score, and fixed setup price. It gives your team a starting point for planning the integration; platform approval and checkout eligibility are separate steps.