The Agentic Commerce Protocol: How Stripe and OpenAI Changed Checkout

In April 2025, Stripe and OpenAI quietly launched something that reshaped online retail: the Agentic Commerce Protocol. It lets AI shopping agents browse a merchant's catalogue, add items to a basket and complete checkout, all without the buyer ever opening a browser. If you sell online in the UK, this is the infrastructure shift you need to understand.

The protocol is already live inside ChatGPT's Instant Checkout feature, and competing implementations are appearing across Google, Amazon and the major card networks. This article explains what the Agentic Commerce Protocol is, how it works, who is building on it, and what UK retailers should do right now to stay visible in the age of agentic commerce.

What is the Agentic Commerce Protocol?

The Agentic Commerce Protocol (commonly shortened to ACP) is a set of standardised APIs that allow AI agents to discover products, read pricing and availability, and complete purchases on behalf of a human buyer. Think of it as a machine-readable shopfront: instead of a customer clicking through pages, an agent calls structured endpoints to do the same thing in seconds.

Before ACP, an AI assistant could recommend a product but had no reliable way to actually buy it. The agent would surface a link, and the human would finish the transaction manually. ACP closes that gap. It gives agents a secure, authenticated path from product discovery through to payment, turning agentic commerce from a recommendation layer into a genuine transaction channel.

Stripe built ACP on top of its existing payments infrastructure, which means any merchant already using Stripe can expose their catalogue to agents with relatively little new engineering. The protocol handles product data retrieval, basket management, payment processing and order confirmation through a single integration layer.

The Stripe and OpenAI partnership

The highest-profile implementation of ACP is Instant Checkout inside ChatGPT. When a user asks ChatGPT to find and purchase a product, the assistant queries ACP-enabled merchants, compares options, and presents a shortlist. The buyer confirms with a single tap, Stripe processes the payment using stored credentials, and the order is placed. No redirect, no form-filling, no abandoned basket.

OpenAI chose Stripe as the payment backbone because of its existing merchant network. Hundreds of thousands of businesses already run on Stripe, which gave ACP immediate scale. For agentic commerce to work, agents need merchants to transact with. Stripe's installed base solved the cold-start problem overnight.

The result is a new kind of checkout experience. The buyer never leaves the conversation. The merchant receives a standard Stripe payment. The agent earns trust by completing the task end-to-end rather than handing off to a web page that might lose the customer. Everyone in the chain benefits, which is why adoption has been rapid.

How ACP works: merchant to agent to buyer

The technical flow is more straightforward than it sounds. A merchant connects their product catalogue to Stripe and enables agent access. Stripe exposes that catalogue through ACP endpoints. When an AI agent receives a shopping request, it queries those endpoints to retrieve product data, titles, descriptions, images, pricing, stock levels, shipping options.

The agent then reasons over the results, applying the buyer's stated preferences and constraints. It might filter by price, compare specifications, or check review sentiment. Once it has a recommendation, it presents the options to the buyer within the conversation interface. If the buyer approves, the agent initiates a Stripe Checkout session through ACP, using the buyer's stored payment method. Stripe handles the transaction, the merchant receives the order, and the buyer gets a confirmation, all within the agent's interface.

Crucially, the buyer's payment details never pass through the AI agent itself. Stripe acts as the trusted intermediary, playing the same role it plays in traditional e-commerce. This separation is what makes ACP viable for regulated markets. The agent orchestrates the purchase, but the financial transaction stays within Stripe's PCI-compliant infrastructure.

Where MCP fits in

If you have been following the AI tools space, you will have encountered another acronym: MCP, or Model Context Protocol. Developed by Anthropic, MCP is a standard that lets AI models connect to external tools and data sources. It is broader than commerce. MCP can connect an agent to a calendar, a database, a code repository, or a product catalogue.

ACP and MCP are complementary rather than competing. MCP provides the general plumbing for agents to interact with external systems. ACP provides the commerce-specific layer on top, covering product schemas, basket logic, payment flows and order tracking. A retailer optimising for agentic commerce benefits from supporting both. MCP makes your data accessible to a wide range of agents. ACP makes your products purchasable through the Stripe-powered ones. Together, they represent the emerging infrastructure of how AI agents choose and transact with merchants.

Who else is building for agentic commerce

Stripe and OpenAI moved first, but they are far from alone. The agentic commerce landscape is filling out rapidly.

Google has been testing its own AI shopping agent, integrated with Google Shopping's existing merchant feeds. The agent uses Google's product data graph (the same infrastructure behind Shopping ads) to match queries to products. For merchants already running Google Shopping campaigns with well-structured feeds, this is a natural extension.

Perplexity launched Perplexity Buy, enabling one-click purchases directly within its AI search results. It positions itself as the answer engine that can also be the checkout, a potent combination for high-intent queries.

Amazon continues to develop Rufus, its conversational shopping assistant embedded across the Amazon ecosystem. Rufus draws on Amazon's vast product catalogue and review database, making it perhaps the most data-rich agentic commerce surface currently in market.

Visa and Mastercard have both announced agent-ready payment rails, designed to let AI agents initiate card-present-equivalent transactions without physical card details changing hands. These network-level integrations mean agentic commerce is not dependent on any single platform. The card networks are building it into the financial infrastructure itself.

The pattern is clear: every major player in search, payments and retail is investing in agent-driven transactions. Agentic commerce is not a niche experiment. It is becoming the standard expectation for how AI-assisted shopping works.

What UK retailers need to do now

The infrastructure is live. The agents are already shopping. The question for UK retailers is whether their products are visible and purchasable when an agent comes looking. Here is what matters most, in order of priority.

1. Get your structured data right

Agents do not browse websites. They query structured data. Your product feeds need to be complete, consistent and machine-readable. Every attribute matters: titles, descriptions, GTINs, pricing, stock levels, dimensions, materials. Missing fields mean missing recommendations. If your feed was built primarily for Google Shopping CPC campaigns, it almost certainly needs reworking for agent consumption.

2. Implement schema markup

Product schema (JSON-LD) on your product pages gives agents another structured signal to work with. Include offers, reviews, availability and shipping information. This is table stakes for agentic commerce visibility. Without it, agents that crawl the web have nothing reliable to parse from your site.

3. Connect to Stripe or an agent-ready payment layer

If you are already on Stripe, investigate enabling ACP access for your catalogue. If you are on another payment provider, check whether they have announced agent-compatible APIs, as most major providers are working on this. The goal is transactability. Agents prefer merchants they can actually buy from over those they can only link to. As our retailer's guide to agentic commerce explains, becoming transactable is the single most impactful change most merchants can make.

4. Optimise descriptions for machine reasoning

Strip the marketing fluff. Agents parse meaning, not persuasion. "Premium artisan hand-crafted excellence" tells an agent nothing. "Organic cotton, 220gsm, GOTS-certified, machine washable at 40C" tells it everything. Specific, verifiable claims win in agentic commerce because agents need facts to match against buyer requirements.

5. Build third-party corroboration

Agents do not take merchants at their word. Products that appear in independent reviews, buying guides and curated marketplace selections carry more weight. Invest in editorial coverage and marketplace presence alongside your direct channel.

How Vendoora handles the hard part

The steps above are clear in principle but demanding in practice. Restructuring feeds, implementing schema, connecting to agent protocols, rewriting product copy for machine consumption: most independent and mid-market retailers do not have the technical resource to do this quickly. Speed matters when the agentic commerce window is opening now.

This is precisely what Vendoora is built to solve. When a retailer joins the Vendoora marketplace, we handle feed structuring, attribute normalisation and schema generation. Your products are listed in a format that AI agents can query, compare and purchase through, without you needing to build your own ACP integration or restructure your existing systems.

Vendoora's commerce layer is already agent-compatible, meaning your products become discoverable and transactable through AI shopping assistants from day one. We also generate the editorial content layer, category guides, product comparisons, expert reviews. That gives agents the third-party corroboration they need to recommend with confidence.

There is no platform fee unless you sell. The model is built around the conviction that agentic commerce is the growth channel for the next decade, and that the retailers who are present and purchasable when agents come shopping will capture disproportionate share.

The protocol era has started

The Agentic Commerce Protocol is not a concept or a roadmap. It is live, processing transactions, and scaling. ChatGPT users are buying products through Instant Checkout today. Google's agent is matching queries to merchant feeds today. Visa and Mastercard are certifying agent-ready rails today.

For UK retailers, the strategic question is no longer whether agentic commerce will matter. It is whether your products will be part of it. The merchants who structure their data, connect to agent-ready payment infrastructure and ensure their products are discoverable and transactable will be the ones agents recommend. Everyone else will be invisible to a growing share of purchasing decisions.

The checkout has moved. Make sure your products have moved with it. UK retailers can start with our practical UK guide, or see how different AI shopping agents use these protocols in practice.

Read next