The next disruption in e-commerce: AI agents in the shopping cart

AI becomes an active participant in commerce

There are no major changes in customer buying behaviour yet, but the signals are clear. Technology companies are investing massively in agentic commerce, and the reason is simple. Online retailers always optimise for conversion; how many clicks and seconds it takes for a purchase to move from thought to payment. Agentic commerce can dramatically accelerate that conversion.

The biggest barrier so far has been that most people don't want to use a purely text-based interface for shopping. But technology has now reached a point where this is no longer a barrier. AI-powered interfaces can operate visually, conversationally and contextually in ways that are beginning to resemble a genuine shopping experience.

When ChatGPT or Claude answers the question "what should I buy as a gift for a 10-year-old who loves Minecraft", it no longer just lists options. AI can now increasingly handle the purchase itself: finding the product, checking availability, comparing prices across stores and completing the order.

Agentic commerce changes what online retailers should focus on developing next.

The purchase journey no longer runs through your website

In traditional e-commerce, the purchase journey begins when a customer arrives at your homepage or product page. The retailer has controlled what the customer sees, in what order, and with what brand voice.

In agentic commerce, the purchase journey begins in AI. The customer asks an agent, which decides which stores to compare. A store that is not accessible to the agent simply isn’t considered.

What matters is understanding what "accessible" means in practice.

An AI agent can in principle navigate any website, but a store interface designed for humans is typically such an awkward environment for an agent that it is not a realistic option. An agent needs a structured interface: something through which it can query products, check availability and proceed to checkout without trying to click and interpret HTML pages.

This introduces a new kind of omnichannel logic: before, you needed to be present on mobile, social channels and search engines. Now you need to be present in AI as well.

MCP: the open standard that enables agentic commerce

Model Context Protocol (MCP) is an open standard developed by Anthropic that defines how AI models can communicate with external systems – such as e-commerce platforms – in a structured and secure way.

MCP acts as an interface between product data and AI. When a store makes its product information, prices, availability and cart management available in MCP format, AI can use them directly by calling clear, structured functions or "tools" or “tools” that the model can call.

This is a meaningful difference compared to a traditional REST API integration or a scraping approach. A REST API tells you what data is available. MCP, in contrast, tells the AI what it can do: fetch a product, check stock, add to cart, and proceed to checkout. The agent doesn't browse but acts.

In practice, MCP allows a store to offer AI surfaces an interface through which an agent can:

• search for products in natural language, without the fragility of scraping

• check availability and price in real time

• add products to the cart and proceed to payment

• apply personalised data, such as purchase history, loyalty benefits and wish lists

The openness of the MCP standard is a critical feature. The same MCP integration works with Claude, ChatGPT, Gemini and any future agent platform. Build it once, reach every AI surface.

The foundation of a future-proof commerce

Building an MCP integration is not a large project on modern e-commerce platforms. It is more of a strategic infrastructure decision: do you add an agent surface to the store's omnichannel strategy now, or wait until competitors have the advantage?

Concretely, an MCP-ready store needs:

1. Product catalogue in MCP format: Product information, descriptions, prices and availability need to be accessible in a structured way – not just as an HTML page. This is often closer to existing capabilities than you might think, if the store already has an API or headless architecture.

2. Functional MCP tools: Cart management, checkout and user authentication need to be offered to agents as clear, directly callable functions. An agent doesn't navigate the DOM, it calls functions.

3. A security and governance layer: The security layer is often the hardest part of the integration. How is the agent identified, and how do you ensure it doesn't carry out actions on the customer's behalf without explicit consent? Guardrails are an essential part of the integration, not an afterthought.

4. Personalisation data for the agent: The biggest competitive advantage comes from the agent being able to use the customer's own data: purchase history, favourites and loyalty status. This requires authentication in an agent context, which is technically solvable.

MCP support may already exist – or be in development – as a built-in feature of modern e-commerce platforms such as Shopify and Commercetools, in which case MCP integration is fairly straightforward to set up. With custom builds, the starting point varies most, but if a headless architecture already exists, the MCP layer is an addition, not a rebuild.

The big players are already moving

This is not a fringe phenomenon in a small startup ecosystem. OpenAI and Google are actively building commerce capabilities directly into their AI models. OpenAI experimented with a direct Instant Checkout feature in ChatGPT, but pivoted away from it in March 2026, shifting focus towards product discovery and merchant-controlled experiences. This is an interesting signal about how the ecosystem is developing: merchants appear to value control over their own customer relationships. Building your own agent surface within your store may be a more natural first step than jumping straight to an external platform.

The standard jointly developed and maintained by OpenAI and Stripe is the Agentic Commerce Protocol (ACP), and Google's equivalent is the Universal Commerce Protocol (UCP). Both are open standards. They handle secure payment facilitation between the agent and the store, but they do not replace MCP. Without the MCP layer, an agent has no structured way to browse the catalogue or manage the cart.

The open standard means a store does not have to commit to a single platform. An MCP integration is a platform-neutral investment — which is why it makes sense to build now, while the agentic commerce ecosystem is still taking shape.

What retailers should do now

The arrival of AI agents in the purchase journey doesn’t mean everything needs to be rebuilt. It means adding one new layer on top of what already exists in two phases.

Phase 1: Build MCP first

The first step is to bring your store's data and functions into MCP format. The result is concrete: an AI-powered shopping assistant in your own store, without external platforms, without commission fees, on the store's own terms. This is a valuable investment regardless of which external standard ultimately wins in the ecosystem.

Concrete steps:

• Assess your current state: does your store already have an API or headless architecture? If so, MCP integration is straightforward to implement.

• Identify priorities: which functions deliver the most value in an agent-driven context? Product search, availability information, cart management?

• Build an MVP integration: publish a narrow product catalogue via MCP and test it with real agent platforms.

• Iterate towards personalisation: once the basic integration works, bring in customer data.

Phase 2: Add ACP/UCP when you're ready

Once the MCP foundation is in place, supporting external agents becomes a much lighter project. ACP and UCP handle secure payment facilitation between external AI platforms and your store. They are the piece that opens your store to users of ChatGPT and Gemini.

The full pipeline then looks like this:

This is a simplified diagram. In practice, ACP does not require MCP. It can also be implemented as a REST interface. MCP and ACP/UCP function as parallel layers: MCP handles product data and the cart, and the ACP/UCP handles the checkout flow and payment.

There is no rush to Phase 2.

The external ecosystem is still evolving, and the situation will become clearer over the coming years. An MCP investment does not go stale while you wait. It delivers value from day one in your own store.

Nitor has the expertise to deliver future-proof MCP-based sales agents your customers can chat with or talk to as well as other tailored solutions.

Conclusion

The history of e-commerce is full of disruptions that seemed small at first, such as mobile shopping, social commerce, and search engine optimisation. Every time, the stores that responded early gained an advantage, and the slower adopters had to play catch-up at great cost.

MCP-based agentic commerce is the next major disruption. It’s technically ready and standardised. This shift will move faster than its predecessors, as AI platforms are developing at an unprecedented pace.

The gateway to future-proof commerce is MCP. It's worth stepping through today.