March 04, 2026

UCP vs ACP: The Two Protocols That Will Decide Your Visibility

Both protocols are live. The full comparison merchants need.

A month ago, we outlined the emerging war between Google and OpenAI over AI-powered commerce. At the time, UCP had just been announced by Shopify and Google at NRF on January 11 and ACP was live but limited.

Since then, UCP has matured quickly. Shopify published detailed engineering documentation, the partner ecosystem expanded to include major retailers, and Klarna and Stripe SPTs were announced. Both protocols are now real, operational, and competing for the same merchants.

This post is the deep-dive. What each protocol actually does, how they differ on the things merchants care about, and why the correct strategy is to support both.

What is UCP?

The Universal Commerce Protocol is an open standard co-developed by Shopify and Google. It defines how AI agents discover products, interact with merchants, and complete purchases - all through a single, interoperable layer.

UCP is built on top of MCP (Model Context Protocol, Anthropic’s open standard for connecting AI models to tools and data). But it goes further by bundling in several specialized sub-protocols:

REST endpoints for basic product data and catalog access
MCP for structured agent-to-merchant communication
Agent Payments Protocol (AP2) for handling money
Agent2Agent (A2A) for agents to coordinate with each other

The launch partners: Walmart, Target, Etsy, Wayfair , and millions of Shopify merchants. Google’s Shopping Graph - 50 billion product listings, 2 billion updates per hour - feeds the discovery layer.

The pitch to merchants: one integration, every AI agent can find you.

What is ACP?

The Agentic Commerce Protocol is OpenAI’s answer, built in partnership with Stripe. It powers Instant Checkout inside ChatGPT , where users can buy products without leaving the conversation.

ACP is open source under Apache 2.0. Merchants host their own product feeds and checkout APIs. The payment infrastructure uses Stripe’s Shared Payment Tokens (SPTs) - single-use, scoped tokens that let AI agents process payments without ever handling raw card data. That’s a meaningful security design: even if an agent is compromised, a stolen SPT is worthless after it expires or gets used once.

The fee: 4% per transaction, collected by OpenAI.

ACP’s first live merchants came from Etsy. Shopify is rolling out support to over a million merchants, including Glossier, Vuori, Spanx, and SKIMS.

The comparison merchants actually need

Forget the protocol specs. Here’s what matters if you run a store:

	UCP (Shopify + Google)	ACP (OpenAI + Stripe)
Backed by	Google, Shopify, Walmart, Target, Etsy, Wayfair	OpenAI, Stripe, Etsy, Shopify
AI surfaces	Google AI Mode, Gemini, any agent using MCP	ChatGPT
Payments	AP2 - supports Visa, Mastercard, PayPal (coming soon), Stripe, Amex, Klarna, Adyen	Stripe Shared Payment Tokens (SPTs)
Merchant fee	Not yet announced	4% per transaction
License	Apache 2.0 open standard	Apache 2.0 open source
BNPL	Klarna integrated	Klarna via Stripe SPTs
What merchants do	Ensure Google Merchant Center feed is current; Shopify merchants get auto-enrolled	Host product feed + checkout API; Shopify merchants get auto-enrolled
EU availability	US-only (designed for global rollout)	US-only (expansion planned)
Data model	Routes through Google Shopping Graph	Merchant-hosted feeds

Two things jump out from this table. First, the partner overlap. Etsy, Shopify, Walmart, Target - they’re in both camps. Second, the payment rails are converging too, with Stripe and Klarna present on both sides.

Everyone is hedging

The most important signal in this launch isn’t which protocol is better. It’s that no major retailer is picking one.

Shopify co-developed UCP with Google while simultaneously rolling out ACP to its merchants. Etsy was the first marketplace live on ChatGPT’s Instant Checkout and also co-developed UCP. Walmart and Target signed up for UCP and have ChatGPT integrations.

This is rational behavior. Both protocols solve different parts of the problem:

UCP gives you reach through Google’s Shopping Graph and any MCP-compatible agent
ACP gives you direct checkout inside ChatGPT, the most popular AI assistant

Picking one means giving up half the potential audience. The big players understood this immediately.

Klarna: the bridge between both worlds

Klarna’s moves on March 3 are worth isolating. The Swedish BNPL giant joined UCP and integrated with Stripe’s Shared Payment Tokens on the same day.

That makes Klarna the first BNPL provider embedded in both ecosystems. For merchants who offer Klarna, this means BNPL works whether a customer buys through a Google AI agent or through ChatGPT. For European merchants watching from the sidelines, Klarna - a European company - bridging both protocols is a signal that this infrastructure is being built with global expansion in mind.

Microsoft is watching (and building)

Microsoft’s Copilot Checkout launched in January and supports both PayPal and Stripe payment rails. Microsoft hasn’t formally endorsed either UCP or ACP, but its payment infrastructure already works with both ecosystems. As we covered in our roundup of AI shopping players, Microsoft is building its own path while staying compatible with the emerging standards.

What merchants should do now

The worst move is to wait and see which protocol wins. Both protocols reward the same thing: complete, structured, machine-readable product data. The work is the same regardless of which agent finds your products.

1. Audit your product data today. If you’re on Shopify, check how your products appear to AI agents right now. Materials, dimensions, weight, care instructions, GTINs - every missing field is a reason for an AI agent to recommend a competitor instead.

2. Keep your Google Merchant Center feed current. UCP reads from Google’s Shopping Graph. Stale prices, missing categories, or out-of-stock items that still show up will hurt you. This was good practice before UCP - now it directly affects your AI visibility.

3. Add GTINs and barcodes to every product. Both protocols use these as canonical product identifiers. Without them, your products are harder to match, verify, and surface. A product without a GTIN is like a book without an ISBN - it exists, but systems can’t reliably find it.

4. Write descriptions for extraction, not persuasion. AI agents don’t care about brand voice. They extract attributes: “waterproof,” “fits true to size,” “dishwasher safe,” “organic cotton.” A product description full of concrete specs beats polished marketing copy in every AI-mediated query.

5. Don’t optimize for one protocol. Both read structured product data. Both reward completeness. The merchant who fills in every attribute wins on UCP and ACP simultaneously.

The European question

Both protocols are US-only today. But the infrastructure is clearly being built for global expansion.

UCP is explicitly designed as an open standard for global adoption - Google’s announcement uses that language directly. Shopify operates globally. Klarna, a Swedish company, is already integrated into both ecosystems. The building blocks for European rollout are there.

ACP’s Instant Checkout currently works for US merchants selling to US buyers. OpenAI has confirmed European expansion is planned but hasn’t given a date. Stripe operates across Europe, so the payment infrastructure isn’t the bottleneck - regulatory approval and merchant onboarding are.

For European merchants, the preparation work is identical to what US merchants should do now. Complete product data. Clean feeds. Structured attributes. The protocols will arrive. The question is whether your catalog will be ready when they do.

The real competition

UCP vs ACP is the headline. But the real competition isn’t between protocols. It’s between merchants who have complete, structured product data and those who don’t.

When a customer asks an AI agent “find me a waterproof hiking jacket under $200 that ships to Portland,” the agent queries both ecosystems. The products that surface are the ones with weight, waterproof rating, materials, shipping zones, and price - all structured, all machine-readable.

The protocol handles the checkout. The data decides the visibility.