Three protocols are going to determine whether your products can be sold through ChatGPT, Google AI Mode, and Perplexity over the next few years. Most retailers I talk to haven't heard of any of them, and that ignorance is going to cost them.
ACP (Agentic Commerce Protocol) is OpenAI and Stripe's standard for product discovery and commerce integration with ChatGPT. UCP (Universal Commerce Protocol) is Google's open standard for purchases inside AI Mode and Gemini, backed by over 60 organizations. MCP (Model Context Protocol), originally from Anthropic and now governed by the Linux Foundation, is the infrastructure layer that lets AI agents connect to any commerce system. Together, these three protocols define how products get discovered, carted, and purchased through AI assistants.
One thing I learned from building commerce infrastructure: when new protocols emerge, the companies that implement early get structural advantages that are very hard to close later.
We're at that moment in commerce right now. So let me break down what ACP, UCP, and MCP actually do, who's behind each one, and what you should be doing about them today.
What Problem Do These Protocols Solve?
Right now, if you want to sell through an AI assistant, there's no standard way to do it. Every platform has built its own approach. ChatGPT has one way of handling product discovery and checkout. Google has another. Perplexity has a third.
For retailers, this creates an ugly situation. Imagine if, in 2005, every search engine required a completely different format for your website. That's roughly where AI commerce is today. Companies like Paz.ai are working on unified integration layers to help retailers avoid building separate implementations for each protocol, but the underlying standards are what make any of this possible.
These three protocols are the industry's attempts to standardize how AI agents discover products, manage carts, process payments, and complete purchases. They come from different companies with different philosophies, and they overlap in some areas while being complementary in others.
What Is ACP (Agentic Commerce Protocol)?
ACP (Agentic Commerce Protocol) is the standard created by OpenAI and Stripe for enabling AI-assisted product discovery, recommendations, and commerce through ChatGPT.
OpenAI and Stripe built ACP, with Shopify as a key launch partner. The protocol was originally designed to let ChatGPT handle the entire checkout process inside the conversation, including payment through Stripe, so the shopper would never visit a website.
In March 2026, OpenAI reversed course. Reporting from The Information revealed that OpenAI discontinued its Instant Checkout feature, moving purchases back to merchant websites and apps. An OpenAI spokesperson said they are "evolving our commerce strategy to better meet merchants and users where they are." Clicking "Buy" in ChatGPT now redirects to the merchant's site or app to complete the purchase.
What still works: ChatGPT shopping is still active and growing. The product discovery and recommendation experience remains strong, with 900 million weekly ChatGPT users (per OpenAI's confirmed figures) getting AI-powered shopping suggestions. ACP as a protocol also still exists and is open-sourced at agenticcommerce.dev. What changed is where the transaction actually happens: on the merchant's site, not inside the chat.
Why this matters: OpenAI discovered that building a full checkout experience inside a chatbot was harder than expected. Payment processing, shipping logistics, returns, customer service disputes, regulatory compliance across jurisdictions - these are hard problems that existing merchant infrastructure already handles well. The pivot validates something important: the real value in AI commerce isn't replacing checkout, it's driving discovery and intent. The infrastructure that connects AI agents to merchant systems (product data, inventory, pricing) matters more than where the "Pay Now" button lives.
Who's integrated with ACP:
- Shopify has ACP enabled across its platform, meaning over a million merchants can surface products in ChatGPT. Purchases redirect to the merchant's Shopify storefront to complete
- Instacart was one of the first ACP partners, with grocery product discovery in ChatGPT directing users to Instacart's app to order
- Target and Etsy have ACP integrations that surface products in ChatGPT conversations, with checkout on their own sites
- PayPal announced ACP integration rolling out in 2026
What it means for you: ACP is now primarily a product discovery and recommendation channel. Your products can appear in ChatGPT shopping conversations, but customers complete purchases on your site. This actually simplifies the retailer's side: you don't need to support a separate anonymous checkout flow. You do need rich product data so ChatGPT can recommend your products confidently, and you need your site's checkout experience to be seamless when that traffic arrives.
What Is UCP (Universal Commerce Protocol)?
UCP (Universal Commerce Protocol) is Google's open standard that enables purchases directly inside Google AI Mode and Gemini search results.
Google built UCP with backing from over 60 organizations including major payment networks, commerce platforms, and global retailers. When someone uses Google's AI search to research a product and the AI shows a "Buy" button right in the results, UCP is the protocol handling that transaction behind the scenes.
The defining feature of UCP is discovery-first commerce. It's designed to work within the context of search and research, turning product exploration into direct purchasing. Google announced UCP supports six core capabilities: product discovery, cart management, identity linking (via OAuth 2.0), checkout, order management, and extensible vertical-specific modules.
Who's using it right now:
- Etsy and Wayfair were the first live UCP partners
- Shopify, Target, and Walmart are in the pipeline for UCP integration (per Google's announcements at their digital advertising events in early 2026)
- Google has also launched "Direct Offers" in AI Mode, which lets merchants surface promotions and deals directly in AI-generated search results
What it means for you: If organic search through Google is a significant channel for your business (and for most retailers, it is), UCP is how you maintain relevance as Google shifts from blue links to AI-generated shopping experiences. Research from Semrush analyzing tens of millions of AI search sessions found that over 90% of AI-generated responses produce zero clicks to external sites. If your products can't be purchased inside AI Mode, they may simply not get purchased at all.
What Is MCP (Model Context Protocol)?
MCP (Model Context Protocol) is a platform-neutral communication standard that enables AI agents to connect to external commerce systems, inventory databases, and payment processors.
Originally created by Anthropic (the company behind Claude), MCP is now donated to the Linux Foundation's Agentic AI Foundation. It was co-founded by Anthropic, Block, and OpenAI, with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg.
MCP is the connectivity layer that lets AI agents talk to external tools and services. It standardizes how AI agents connect to external services the way REST APIs standardized web service communication a decade ago. It doesn't handle the shopping experience directly, but it provides the plumbing that lets an AI agent connect to your inventory system, check real-time pricing, verify stock levels, or process a return.
Where ACP and UCP handle the consumer-facing shopping experience, MCP operates at the infrastructure level. It's what enables an AI agent to pull your product catalog, check whether an item is in stock at a specific warehouse, or initiate a payment through a connected service.
The defining feature is that MCP is platform-neutral. Because it sits at the Linux Foundation, no single company controls it. MCP adoption is accelerating rapidly, with major cloud providers (AWS, Google Cloud, Azure) all offering managed MCP hosting.
Who's using it right now:
- Visa deployed MCP tools on AWS as part of their Intelligent Commerce partnership, handling identity verification, fraud detection, and settlement
- Worldpay launched MCP servers for payment integration with AI agents
- Google launched managed MCP servers for Maps, BigQuery, and other services (TechCrunch described it as Google going "all-in on MCP")
- Shopify stores can connect to AI agents through MCP for inventory, pricing, and fulfillment data
What it means for you: MCP is the foundational layer. Even if you're not implementing ACP or UCP today, ensuring your commerce infrastructure can communicate with AI agents through MCP is how you stay accessible as the ecosystem evolves. It's also particularly relevant for B2B. Gartner predicts that 90% of B2B buying (over $15 trillion in spend) will be AI agent-intermediated by 2028, and MCP is the protocol that makes that machine-to-machine commerce possible.
How the Three Protocols Work Together
These aren't competing standards in the way that VHS competed with Betamax. They operate at different layers:
Consumer-facing discovery + commerce: ACP (ChatGPT) | UCP (Google)
Infrastructure/connectivity: MCP (universal)
MCP is the foundation. ACP and UCP sit on top, handling commerce-specific workflows for their respective platforms. A retailer that wants full coverage across AI shopping channels will eventually need presence in both ACP and UCP ecosystems, with MCP providing the data connectivity underneath. Platforms like Paz.ai are building unified integration layers across all three protocols so retailers don't have to manage each one separately.
It's worth noting that ACP and UCP are evolving differently. UCP is pushing toward full in-platform checkout within Google AI Mode. ACP, after OpenAI's March 2026 pivot, now focuses on product discovery and recommendation within ChatGPT while routing purchases to merchant sites. This difference actually reinforces why MCP matters most at the infrastructure level: regardless of where checkout happens, the AI agent still needs to access your product data, inventory, and pricing.
PayPal's approach is instructive here. They're supporting all three: ACP integration for ChatGPT, UCP membership in the Google coalition, and their own MCP Agent Toolkit for developers. Major payment processors are hedging across all protocols because they expect all three to matter.
Which Protocol Matters Most for Your Platform?
Your existing commerce platform shapes which protocol you should prioritize:
If you're on Shopify: You're in the best position right now. Shopify has integrated with ACP (giving you ChatGPT product discovery), is joining UCP (Google AI Mode access), and supports MCP connections. Your main action item is ensuring your product data is rich enough to perform well when AI agents evaluate it. Shopify's "agentic storefronts" feature makes your store accessible through ChatGPT, Copilot, and Gemini.
If you're on Magento/Adobe Commerce: Adobe is part of the UCP coalition, so Google AI Mode integration should come through that channel. ACP integration will likely require middleware or a third-party connector. Focus on MCP readiness through your product data feeds and APIs, and ensure your Merchant Center feed is complete.
If you're on Salesforce Commerce Cloud (SFCC): Salesforce offers ACP support through its agent commerce capabilities, which gives you a path to ChatGPT product visibility. UCP integration will depend on Salesforce's rollout timeline. The platform also positions SFCC for MCP-based agent interactions.
If you're on a custom or headless platform: You have the most flexibility but also the most work. You'll need to build protocol integrations yourself or through partners like Paz.ai. The upside is you can implement exactly what you need. Start with MCP (broadest applicability) and add ACP or UCP based on where your customers are shopping with AI.
How Should Retailers Prepare for AI Commerce Protocols?
You don't need to implement all three protocols tomorrow. But you do need to start preparing, because the retailers moving now will have meaningful advantages by the time these protocols reach mainstream adoption.
This week:
Check your Google Merchant Center feed. Make sure it's complete, accurate, and includes the
native_commerceattribute that UCP will require. If you don't have a Merchant Center account, set one up. This is table stakes for UCP.Whitelist AI crawlers. Check your robots.txt for GPTBot, ClaudeBot, and PerplexityBot. If they're blocked, AI agents can't index your products. This takes 15 minutes.
Audit your product data depth. AI agents make recommendations based on how completely they understand your products. Pull up your top sellers and ask: does this product listing contain enough detail for an AI to recommend it with confidence over a competitor? If not, start enriching.
This month:
Join the waitlists. Google's UCP has early access programs. OpenAI's ACP program is open to merchants who want their products surfaced in ChatGPT shopping. Getting on them now means you'll have integration support when you're ready to implement.
If you're on Shopify, activate agentic features. Shopify has made agent-readiness a default capability. Make sure it's turned on and that your product data is optimized for conversational queries, not just keyword searches.
Assess your API readiness for MCP. Can an external system query your product catalog, check inventory, and get pricing through an API? If not, that's your infrastructure gap. MCP connectivity depends on having accessible, well-documented APIs.
This quarter:
Run a protocol readiness assessment. Map your current capabilities against what each protocol requires. Identify the gaps. Build a roadmap. Paz.ai offers protocol readiness assessments for retailers evaluating their AI commerce strategy. The retailers who treat this as a strategic initiative (rather than a tactical IT project) will move faster.
Start measuring AI visibility. Track whether your products appear in ChatGPT, Perplexity, and Google AI Mode for relevant queries. This is the baseline you'll improve against as you implement protocol support.
Frequently Asked Questions
What is the difference between ACP and UCP?
ACP (Agentic Commerce Protocol) powers product discovery and shopping recommendations within ChatGPT, built by OpenAI and Stripe. UCP (Universal Commerce Protocol) handles purchases within Google AI Mode and Gemini, backed by Google and 60+ organizations. After OpenAI discontinued Instant Checkout in March 2026, the key difference is that UCP aims to keep the full transaction inside Google's platform, while ACP now focuses on discovery and recommendations in ChatGPT before redirecting to merchant sites for checkout. Both protocols solve the same underlying problem - connecting AI assistants to commerce - but they take different approaches to where the purchase actually completes.
Do I need all three protocols?
MCP is the foundational layer that all AI agents will use for data connectivity. ACP and UCP are channel-specific. You need ACP if your customers use ChatGPT for shopping discovery, UCP if they use Google AI Mode. Most major retailers will eventually implement all three, but start with whichever aligns with your highest-traffic AI channel and build MCP connectivity as your infrastructure baseline.
How do I get started with AI commerce protocols?
Start with the basics: ensure your Google Merchant Center feed is complete, whitelist AI crawlers in your robots.txt, and audit your product data for depth and accuracy. Then join the early access programs for ACP and UCP. If you want to move faster, platforms like Paz.ai provide multi-protocol integration so you don't have to build each connection from scratch.
What does MCP mean for B2B commerce?
MCP is especially significant for B2B because it enables AI agents to access complex product catalogs, negotiate pricing, check inventory across warehouses, and manage procurement workflows. Gartner predicts that 90% of B2B buying will be AI agent-intermediated by 2028, and MCP is the protocol that makes that machine-to-machine commerce possible.
Did OpenAI kill AI shopping in ChatGPT?
No. OpenAI discontinued the Instant Checkout feature (where you could complete a purchase without leaving ChatGPT), but ChatGPT shopping is still very much alive. Product discovery, recommendations, and comparison shopping all still work through ACP. The change is that when you're ready to buy, you're redirected to the merchant's website or app. Think of it like how Google Shopping shows you products and prices but sends you to the retailer to buy. The discovery layer is often the most valuable part of the funnel anyway.
Why Early Protocol Adoption Matters
Morgan Stanley research suggests roughly a quarter of American consumers have made AI-assisted purchases. That's not early adopters anymore. Industry studies indicate a majority of consumers now use AI tools at some point in their shopping journey. The question for retailers has shifted from "will AI shopping matter?" to "will we be ready when our customers expect it?"
The protocol landscape will continue evolving. OpenAI's March 2026 decision to move checkout back to merchant sites is a perfect example. New capabilities will be added, standards will converge in some areas, and the technical requirements will change. But the retailers who build their foundation now (clean data, protocol awareness, API readiness) will adapt to those changes much more easily than those starting from scratch.
I saw this dynamic play out with OAuth adoption in payments. The companies that implemented early influenced how the spec evolved and built integrations that became the reference implementations. The ones that waited inherited someone else's decisions and spent twice as long catching up. The same pattern is forming around ACP, UCP, and MCP right now.
Dor Shany is the CEO of Paz.ai, an agentic commerce platform that helps retailers sell through AI shopping agents. This article reflects his analysis of publicly available information. More at paz.ai/blog.
Top comments (0)