A Retailer’s Guide to AI Shopping Protocols: ACP, UCP, and MCP Explained

Three protocols now determine whether your products appear when a shopper asks an AI assistant for a recommendation. Most retailers have not heard of any of them. That gap is closing fast, and the brands that close it first will own a distribution advantage that compounds for years.

The Fragmentation Problem Nobody Warned You About

AI-referred traffic to Shopify grew 7x between January 2025 and early 2026. That number is not a projection. It is a measured outcome from a channel most retailers were not actively building for. The brands showing up in those results are not there by accident. They are there because they understood, earlier than most, that AI shopping runs on infrastructure, and infrastructure requires deliberate preparation.

The problem is that there is no single standard for selling through an AI assistant. ChatGPT handles product discovery differently from Google AI Mode, which handles it differently from Perplexity. For retailers, this creates the same fragmentation that early web commerce faced before payment and search standards emerged. You cannot build once and distribute everywhere, at least not yet, because the protocols governing each platform are at different stages of adoption and serve different layers of the transaction.

MCP, ACP, and UCP are the industry’s coordinated response to that fragmentation, each operating at a distinct layer of the stack. Understanding what each one does, where it sits, and how they interact is now foundational knowledge for any retailer serious about AI-driven commerce. This is not theoretical infrastructure planning. As of March 2026, Shopify has activated Agentic Storefronts by default for all eligible US merchants, Google announced UCP at NRF in January 2026 with Sundar Pichai on stage, and Stripe launched its Agentic Commerce Suite in December 2025 with URBN, Etsy, Ashley Furniture, Coach, Kate Spade, and Revolve already onboarding. The infrastructure is live. The question is whether your catalog is ready to be found inside it.

MCP: The Data Plumbing Layer Every AI Agent Runs On

MCP (Model Context Protocol) is the foundation of the entire stack. Anthropic open-sourced it in November 2024, and within six months Google, Microsoft, OpenAI, Visa, and Mastercard had all committed to the standard. Shopify activated a default MCP endpoint on every store on the platform as part of the Summer 2025 Edition, meaning your store is already reachable by AI agents if you are on Shopify. The question is not whether you have MCP. The question is whether the data behind your endpoint is complete enough for an agent to act on it confidently.

The protocol follows a client-server architecture. The MCP server lives on the merchant side, exposing structured data about products, inventory, pricing, policies, and checkout flows. The MCP client lives inside the AI assistant, reading that data and acting on it in real time. When a shopper tells ChatGPT “find me a waterproof hiking boot under $180 with wide sizing,” the AI client queries MCP servers across eligible merchants, compares results against the shopper’s constraints, and surfaces the best matches with an embedded checkout option, all inside the conversation. That entire process happens without the shopper visiting a single product page.

Think of MCP as what APIs were to mobile apps. Before standardized APIs, every mobile application had to build custom integrations with every external service it needed to access. MCP solves the same problem for AI agents and commerce systems. Without it, an AI agent has no reliable way to access real-time product information, inventory status, pricing, or policy data from your systems. With it, any compatible agent can query your catalog directly, without a custom integration built for each platform. Walmart’s “Super Agent” architecture illustrates this at scale: three distinct agent personas (Sparky for customers, Associate for employees, Marty for suppliers) share a single MCP-powered data plane. When Walmart updates inventory, all three agents read from the same source automatically. That consistency is what makes agent commerce trustworthy.

For retailers not on Shopify, MCP readiness means ensuring your product data feeds and APIs are structured, current, and accessible. The specific implementation path varies by platform, but the underlying requirement is consistent: clean, complete, machine-readable data. You can review the complete guide to agentic commerce for Shopify merchants for a detailed breakdown of how Shopify’s MCP infrastructure works alongside Catalog and Agentic Storefronts.

ACP: The ChatGPT Commerce Layer and What Actually Happened to Instant Checkout

ACP (Agentic Commerce Protocol) is the open standard co-developed by OpenAI and Stripe that connects your product catalog to ChatGPT’s 700 million weekly users. It is open-source under the Apache 2.0 license and community-designed, meaning any business can implement the specification to transact with any compatible AI agent using any compatible payment provider. OpenAI is the first platform to implement ACP, but the specification is not locked to ChatGPT.

Here is the architecture as it works today: your business publishes an ACP-compatible checkout endpoint. When a shopper inside ChatGPT selects a product and confirms a purchase, the agent makes a request to your business to initiate checkout on your behalf. Your business receives the request along with a Shared Payment Token (SPT), a new payment primitive that allows the AI agent to initiate payments using the buyer’s saved payment method without exposing underlying credentials. Every SPT can be scoped to a specific seller, bounded by time and amount, and monitored throughout its lifecycle. You remain the merchant of record. You control which products can be sold, how they are presented, how transactions are processed, and how orders are fulfilled.

In September 2025, OpenAI launched “Instant Checkout” inside ChatGPT, allowing purchases to complete entirely within the chat interface. By early 2026, they had walked it back. The official statement: “Instant Checkout is moving to Apps, where purchases can happen more seamlessly.” The real reason was more direct: brands rejected being reduced to anonymous fulfillment centers. When a purchase completes inside ChatGPT without ever touching the merchant’s storefront, the retailer loses the traffic, the customer data, the login and registration flow, and the loyalty engagement that makes a second purchase more likely. Shopify president Harley Finkelstein made this explicit at a Morgan Stanley conference in March 2026: the checkout itself, the subscriptions, the inventory, the shipping and taxes, the merchandising options, those are the moat. ACP today is a product discovery and catalog protocol. The transaction redirects to your storefront. You keep the customer relationship.

In December 2025, Stripe launched its Agentic Commerce Suite to reduce the integration burden further. Rather than building and maintaining your own ACP endpoints, you connect your product catalog to Stripe, select which AI agents you want to sell through in the Stripe Dashboard, and Stripe handles discovery, checkout, payments, and fraud detection. Leading brands including URBN (Anthropologie, Free People, Urban Outfitters), Etsy, Ashley Furniture, Coach, Kate Spade, and Revolve are already onboarding. The Suite is also rolling out through ecommerce platforms including Wix, WooCommerce, BigCommerce, Squarespace, and commercetools. If you use Stripe for payments today, enabling agentic payments requires updating as little as one line of code. For more on how ChatGPT commerce works specifically for Shopify brands, the ChatGPT commerce for Shopify guide covers the setup steps and measurement approach in detail.

UCP: The Google-Native Standard Built for Commerce Complexity

UCP (Universal Commerce Protocol) is the most architecturally ambitious of the three. Co-developed by Shopify and Google and announced by Sundar Pichai at NRF in January 2026, it is designed to handle the full complexity of real retail commerce across any AI surface that adopts it. The coalition behind it includes Walmart, Target, Best Buy, The Home Depot, Wayfair, Etsy, Mastercard, Visa, PayPal, American Express, and Zalando, along with 20-plus additional organizations. Where ACP is currently tied primarily to the ChatGPT ecosystem, UCP is explicitly platform-agnostic.

The technical design reflects hard-won knowledge about what actually breaks in real commerce. Shopify spent over 20 years learning that payment options differ based on cart contents, buyer location, and market; discounts have stacking and combination rules that rival the tax code; fulfillment options multiply into hundreds of permutations. UCP models this complexity rather than flattening it. Merchants publish a capability profile declaring what they support, including their own bespoke functionality. Agents discover these capabilities, negotiate what they can handle, and proceed to complete transactions. The protocol is layered: a Shopping Service defines core transaction primitives, Capabilities add major functional areas (Checkout, Orders, Catalog), and Extensions allow domain-specific schemas to be added without breaking the core spec.

The Embedded Checkout Protocol (ECP), a component within UCP, handles the cases where human input is genuinely required during checkout. A furniture retailer that needs the customer to select a specific delivery date and time can specify exactly what information is needed. ECP establishes a bidirectional JSON-RPC 2.0 channel between agent and merchant, with checkout embedded in the agent’s surface. The buyer’s native payment sheet surfaces for payment collection. Address selection pulls from the agent’s wallet. The merchant gets structured data to finalize the transaction. The buyer never leaves the conversation, but the merchant’s business logic is preserved intact.

Payment negotiation in UCP works dynamically. Rather than the protocol defining every payment method, each provider (Google, Shopify, or any regional PSP) publishes its own handler specification. The merchant advertises which handlers it accepts. The agent picks one and follows its spec. Change the cart contents, the buyer’s region, or any other variable, and the available handlers may shift. New payment methods enter the ecosystem without requiring committee votes or core version changes. This is the architecture Google is using to power the buy button now appearing directly in Google Search AI Mode and the Gemini app.

How the Three Protocols Work Together

These are not competing standards. They operate at different layers and are designed to be complementary. MCP provides the data connectivity. ACP and UCP sit on top, handling the commerce-specific workflows for their respective platforms. A retailer wanting full coverage across AI shopping channels will need presence in both the ACP and UCP ecosystems, with MCP providing the data layer underneath both.

UCP is explicitly designed to be compatible with MCP, as well as Google’s Agent2Agent (A2A) protocol and the Agent Payments Protocol (AP2). Sundar Pichai confirmed this at NRF: “UCP is compatible with existing industry protocols like Agent2Agent, the Agent Payments Protocol, and Model Context Protocol.” The payment networks have reached the same conclusion. PayPal adopted ACP in October 2025 to power in-chat payments within ChatGPT. PayPal is also part of the UCP coalition. Mastercard’s Chief Digital Officer confirmed participation across UCP, A2A, AP2, and ACP simultaneously. When every major payment network is hedging across all three protocols because they expect all three to matter, the infrastructure signal is unambiguous.

The pattern that has emerged from conversations with hundreds of operators is consistent: the retailers winning early in this environment are not treating protocol adoption as a technical project. They are treating it as a data and policy problem. The protocols are the plumbing. What flows through the plumbing is your product catalog, your pricing, your inventory signals, your return policies, your fulfillment timelines. If that data is thin, inconsistent, or incomplete, the protocols cannot help you. If that data is clean, structured, and current, the protocols distribute it to every AI surface that adopts them. That is the leverage. For a real-time look at what happened the week this infrastructure went fully live, the week agentic commerce stopped being theoretical covers the coordinated moves by Shopify, OpenAI, Google, and Walmart that confirmed the architecture.

What This Means for Your Platform Today

The practical implications depend on where you are operating today, and they vary significantly by platform.

Shopify merchants are in the strongest position of any platform. Shopify co-developed UCP with Google and integrated with ACP. As of March 24, 2026, Agentic Storefronts are live by default for all eligible US merchants. Your products are likely already appearing in ChatGPT, Microsoft Copilot, and Google AI Mode. Your primary action item is not protocol integration. It is ensuring your product data is complete enough for AI agents to recommend your products with confidence. Navigate to Settings, then Sales Channels in your Shopify Admin to confirm your Agentic Storefronts status and verify which AI platforms are enabled. The Shopify Agentic Storefronts setup guide walks through the full verification and catalog audit process step by step.

Adobe Commerce merchants have a clear path through UCP. Adobe is part of the UCP coalition, so Google AI Mode integration should come through that channel as the protocol rolls out. ACP integration will likely require middleware or a third-party connector in the near term. In the meantime, focus on MCP readiness through your product data feeds and APIs, and ensure your Google Merchant Center feed is complete and accurate. Clean Merchant Center data is the prerequisite for everything that follows.

Salesforce Commerce Cloud merchants can start with ACP. Salesforce launched Agentforce Commerce with ACP support, providing a path to ChatGPT product discovery. UCP integration will depend on Salesforce’s rollout timeline. The Agentforce platform also positions SFCC for MCP-based agent interactions. Watch Salesforce’s roadmap communications closely through mid-2026, as this is where the most meaningful platform-level changes are likely to surface.

Custom and headless platform merchants have the most flexibility and the most work. You will need to build protocol integrations yourself or through partners. Start with MCP, which has the broadest applicability across AI platforms, then layer in ACP or UCP based on where your customer research tells you your buyers are actually using AI to shop. Do not build for all three simultaneously. Sequence by channel priority and measure before expanding.

The Real Bottleneck Is Not the Protocols

The conversation I keep having with retailers is this: they want to talk about protocol compliance, and I want to talk about their product data. The protocols are the distribution layer. What they distribute is your catalog. And most catalogs are not ready.

A typical retail product listing has five to eight structured attributes. AI agents need thirty or more to make a confident recommendation. When 70% of ChatGPT shopping queries include specific constraints like “waterproof,” “under $200,” or “vegan,” the agent is filtering by attributes. If your catalog does not have those attributes structured and machine-readable, you are filtered out before you are ever considered. It does not matter which protocols you support if the data behind them is thin.

AI agents do not browse product pages the way humans do. They do not respond to lifestyle photography or clever taglines. They read structured attributes and make matching decisions based on how precisely your product data answers a shopper’s stated constraints. When someone tells an AI agent “I need a moisturizer for oily skin, fragrance-free, under $40, ships in two days,” the agent queries your endpoint and evaluates whether your product data contains clear answers to every one of those constraints. If it does, you are a candidate. If it does not, you are invisible. This is not a future risk. It is the current state of how AI shopping works.

You can use Paz.ai to audit and optimize your catalog for AI agent discoverability, and see exactly where your products stand against competitors across ChatGPT, Perplexity, and Google AI. The tool surfaces the specific attribute gaps that are costing you recommendations, which gives you a prioritized list of data improvements rather than a generic instruction to “add more attributes.”

Where to Start This Week

The sequence matters. Do not start with protocol applications. Start with the data and access prerequisites that make those applications worth submitting.

First, audit your Google Merchant Center feed. UCP requires structured, consistent product data. If your feed has gaps in attributes, sizing, materials, or compatibility information, AI agents will recommend your competitors instead. If you do not have a Merchant Center account, creating one is the first step. The feed is the foundation that UCP, ACP, and MCP all draw from in different ways. Getting it clean is not platform-specific work. It is work that benefits every channel simultaneously.

Second, check your robots.txt for GPTBot, ClaudeBot, and PerplexityBot. If any of these are blocked, AI agents cannot index your products. This is a 15-minute fix with meaningful consequences. Add explicit Allow directives for each crawler and verify the change with a crawler testing tool before moving on.

Third, evaluate your product data depth on your top 20 SKUs by revenue. Pull each listing and ask: does this contain enough structured detail for an AI to recommend it confidently over a competitor? Not marketing copy. Structured attributes. Materials, dimensions, use cases, compatibility, care instructions, return windows, shipping timelines. If the answer is no for more than half your top products, that is your real starting point and the work that will produce the most visible results fastest.

Once those foundations are solid, apply for early access to ACP through chatgpt.com/merchants and to Google’s UCP Merchant Center onboarding. If you are on Shopify, verify your Agentic Storefronts settings and confirm which AI platforms you are opted into. Then start measuring: search for your own products in ChatGPT, Perplexity, and Google AI Mode using the queries your customers actually use. That baseline is what you will improve against, and it will tell you more about your current AI commerce readiness than any protocol audit ever could.

Frequently Asked Questions

What is the difference between MCP, ACP, and UCP in plain terms?

MCP (Model Context Protocol) is the data connectivity layer that lets AI agents read your product catalog, inventory, pricing, and policies in real time. ACP (Agentic Commerce Protocol) is the transaction standard co-developed by OpenAI and Stripe that governs how purchases happen through ChatGPT. UCP (Universal Commerce Protocol) is the broader commerce standard co-developed by Shopify and Google that handles the full complexity of checkout (discounts, loyalty, subscriptions, fulfillment options) across Google AI Mode, Gemini, and any platform that adopts it. MCP is the foundation. ACP and UCP are the commerce layers built on top of it for their respective ecosystems. A retailer wanting full AI channel coverage will eventually need all three.

Do I need to implement all three protocols to sell through AI platforms?

Not immediately, and not all at once. If you are on Shopify, MCP and UCP are already active through Agentic Storefronts, and ACP integration is available through your Shopify Admin. If you are on Adobe Commerce or Salesforce, your fastest path is through UCP via your platform’s coalition membership, with ACP requiring additional middleware in the near term. Custom and headless platforms need to build integrations directly. The practical recommendation is to sequence by where your customers are actually shopping with AI, start with one platform, measure, then expand. Do not attempt to build for all three simultaneously if you have limited technical resources.

What happened to Instant Checkout inside ChatGPT and does ACP still matter?

OpenAI launched Instant Checkout in September 2025, allowing purchases to complete entirely inside the ChatGPT conversation. By early 2026, they shut it down, with the official statement being that it was “moving to Apps.” The real driver was that brands rejected the model: when purchases complete inside ChatGPT, retailers lose the traffic, customer data, loyalty flows, and post-purchase relationship that make repeat purchases possible. ACP absolutely still matters. It is now the infrastructure connecting ChatGPT’s product discovery to merchant storefronts, with the purchase completing on the retailer’s site. You keep the customer relationship. The protocol is more valuable in this architecture than in the original in-chat checkout model.

How do I know if my Shopify store is already visible in AI shopping results?

Start in your Shopify Admin under Settings, then Sales Channels, and look for the Agentic Storefronts section. Confirm the channel is active and that ChatGPT, Microsoft Copilot, and Google AI Mode are all toggled on. Then run the actual test: open ChatGPT, Perplexity, and Google AI Mode and search for the products you sell using the specific language your customers use, including constraints like price, material, use case, and size. If your products appear, your data is working. If they do not, or if competitors consistently appear instead, you have an attribute gap in your catalog data. That gap is what to fix first, before any protocol application.

What product data attributes do AI agents actually need to recommend my products?

AI agents filter by the specific constraints shoppers state in their queries. Research on ChatGPT shopping queries shows that 70% include at least one specific constraint such as price range, material, use case, or compatibility requirement. Your product listings need to answer those constraints with structured, machine-readable attributes, not marketing copy buried in a description paragraph. For most categories, this means: materials and composition, dimensions and sizing with fit notes, use cases and compatibility, care instructions, return window, shipping timeline, and any certifications relevant to your category (vegan, organic, waterproof rating, etc.). Pull your top 20 products by revenue and audit each one against the question: if a shopper stated five specific constraints, can an AI agent confirm a match from this listing alone?