AP2 vs ACP vs MCP vs UCP: Which Agentic Commerce Protocol Should You Support?

TL;DR

• Most merchants will need more than one agentic commerce protocol, because each covers a different layer of the transaction.
• AP2 proves an agent was authorized to pay, ACP standardizes checkout for agents in ChatGPT, MCP is the connection agents use to reach catalog and cart tools, and UCP runs discovery through checkout in Google AI Mode and Gemini.
• Start with the groundwork every protocol reads (structured product data, a liability policy and measurement), then add the integration that matches where your buyers shop and treat it as reversible.
• Promptwatch's prompt tracking runs your buyer-intent prompts across ChatGPT, Google AI Mode and other AI platforms, so you can see which surface your buyers already use before you commit integration work to one protocol.

Why “picking a protocol” is the wrong question in 2026

Every comparison of AP2, ACP, MCP and UCP is implicitly built around a “which one wins” frame, and that frame is a poor basis for a 2026 roadmap. Each of the four protocols handles a different part of an agentic commerce transaction:

• AP2 proves an agent had real authorization to pay
• ACP lets a business become purchasable directly inside a chat surface
• MCP connects an agent to a merchant's catalog and cart tools
• UCP covers discovery, cart and checkout in one specification and runs checkout in Google AI Mode and Gemini

A merchant asking “should I build for AP2, ACP, MCP or UCP” is often asking four separate questions at once, and answering them as one is how a roadmap gets stuck on a bet that doesn't pay off.

ACP is the clearest evidence this happened already. When ChatGPT's Instant Checkout launched in late 2025:

• Etsy went live immediately
• Stripe named Glossier, Vuori, Spanx, and SKIMS as Shopify merchants coming soon
• The pitch was a “single integration” that would let a business “start selling through AI agents while retaining full control over what's sold”

By March 2026, OpenAI had refocused ChatGPT on product discovery and moved checkout to merchants' own connected apps, after real enablement lagged well behind the pitch. A merchant that treated ACP integration as its whole readiness project had to rework that bet within months.

What all four protocols actually require from merchants

Underneath the branding, all four protocols need the same foundation before they need anything protocol-specific:

• Complete, machine-readable product data: price, real-time availability, shipping terms, return policy
• A written position on what happens when an agent gets a purchase wrong

AP2's Mandates, ACP's Checkout API, MCP's catalog tools and UCP's checkout capability all read that same underlying data; none of them create it for you. That's the groundwork that pays off no matter which protocol, or combination, ends up mattering for your buyers. For the fuller readiness checklist, structured data, a liability policy, and measurement, see how to prepare for agentic commerce.

What changes if a protocol wins vs. loses adoption

If a protocol you bet on loses adoption, the work at risk is the protocol-specific integration you built around it: custom checkout logic, a single-vendor payment flow, an API contract that assumed that protocol was the only one you'd ever need. Structured product data, a liability policy and measurement infrastructure keep their value, since none of them is protocol-specific. Keep the reversible layer (protocol integration) small and put the real investment in the durable layer (readiness), so a scaled-back protocol only costs you the integration work.

AP2 (Google-led): what it is and who's backing it

The Agent Payments Protocol standardizes proof of authorization through two cryptographically signed “Mandates”:

• An Intent Mandate capturing what the user actually asked for
• A Cart Mandate locking the final price before payment

Together they create an audit trail showing an agent had real permission from the user to spend. Google published AP2 in September 2025 with a broad, named partner list across payments and enterprise software, among 60+ organizations:

• Mastercard, PayPal, American Express
• Coinbase, Adobe, Salesforce

Payment support is wide by design:

• Cards, digital wallets, bank transfers
• Stablecoins and other crypto payments, through a separate A2A x402 extension built with Coinbase, the Ethereum Foundation, and MetaMask

AP2's own announcement names no merchant deployment. That is expected for this layer: retailers integrate payment-authorization infrastructure through a processor or platform and rarely announce it individually. It does mean “AP2 has the deepest backer bench” and “AP2 has proven retail adoption” are two different claims, and only the first is currently verifiable from Google's own documentation.

ACP (OpenAI/Stripe-led): what it is and who's backing it

The Agentic Commerce Protocol standardizes a Checkout API and a Delegate Payment API that let a business become purchasable directly inside a chat surface, using tokenized payment credentials so the agent never touches a raw card number. OpenAI and Stripe are the joint maintainers, with governance moving toward a neutral foundation. Payment handling is more flexible than the “Stripe protocol” label suggests: a merchant can process through Stripe directly, or hand payment to another provider entirely while still using Stripe's fraud risk scores.

ACP's adoption story is the most public of the four, for better and worse:

• At launch (late 2025): Etsy went live with Instant Checkout; Stripe named Glossier, Vuori, Spanx, and SKIMS as Shopify merchants coming soon, pitched as a “single integration.”
• By March 2026: OpenAI refocused ChatGPT on product discovery and moved checkout to merchants' connected apps, after real merchant enablement fell short. One industry analysis put actual Instant Checkout adoption at roughly a dozen Shopify merchants against that launch framing (a secondary-source figure that OpenAI and Stripe haven't confirmed, so treat it as directional).

The protocol itself didn't disappear; it continues to be used outside that specific chat surface, and its own GitHub repository and Stripe's documentation remain the primary technical reference. But the gap between the launch pitch and what merchants actually enabled is the single most concrete data point in this entire comparison, and it's the reason “ACP has the most name recognition” and “ACP has the most retailer adoption” are no longer the same statement.

MCP-based commerce extensions: what they add

The Model Context Protocol, created by Anthropic as a general-purpose way for an AI system to connect to external data and tools, isn't commerce-specific on its own. Commerce platforms give it a commerce role by exposing catalog, cart and checkout tools through an MCP server. Shopify shows where this is heading: it deprecated its standalone Storefront MCP catalog and cart tools in June 2026, set August 31, 2026 as the removal deadline, and moved them to a UCP endpoint at https://{shop}/api/ucp/mcp. On Shopify, MCP is now the transport and UCP defines the commerce capabilities.

MCP leaves payment authorization to other layers. It carries the tool calls, so the merchant's own checkout, UCP's checkout capability or a payment-authorization layer like AP2 handles the actual transaction. That narrower scope is why merchants usually pair MCP with another protocol, and it's why UCP uses MCP as one of its transports.

UCP (Google and Shopify-led): what it is and who's backing it

The Universal Commerce Protocol is an open standard that covers the full agent shopping journey in one specification: product discovery, cart, checkout and order management. Google published it on January 11, 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. More than 20 companies endorsed it at launch, including Adyen, American Express, Best Buy, Flipkart, Macy's, Mastercard, Stripe, The Home Depot, Visa and Zalando, according to the Google Developers Blog.

UCP works through three building blocks:

• Capabilities: standard commerce functions such as catalog search, cart, checkout, order updates and identity linking
• Extensions: optional add-ons on top of a capability, such as discounts, fulfillment options, loyalty and AP2 Mandates
• A discovery profile: each business publishes what it supports at /.well-known/ucp, and the agent negotiates the features both sides share before it acts

UCP builds on two of the other protocols in this comparison. MCP is one of its supported transports, alongside REST and Agent2Agent (A2A). AP2 plugs in as a checkout extension that carries cryptographic proof of what the user authorized. The merchant stays merchant of record and keeps control of its own checkout logic.

The specification is moving fast. A March 19, 2026 update added three things:

• Multi-item carts
• Real-time catalog access to variants, inventory and pricing
• Identity linking, so loyalty perks carry across platforms

The same update named Commerce Inc, Salesforce and Stripe as platforms planning to implement UCP. The spec is versioned by date, and the current version at the time of writing is 2026-04-08.

What it takes to sell through UCP on Google today

Google's own implementation of UCP powers checkout in AI Mode in Google Search and in the Gemini app. As of October 2026, Google's Merchant Center documentation lists these conditions:

• The merchant is based in the United States, Canada or Australia
• An active Merchant Center account with current product data, and the native_commerce checkout eligibility attribute set on listings
• A Google Pay & Wallet Console account, with your payment provider onboarded to the Google Pay API
• An interest form, since UCP checkout is still in early access
• Shoppers currently pay with cards stored in Google Wallet

That ties UCP closely to a marketing asset most retailers already maintain: the Merchant Center feed. A brand that keeps its Google Shopping feed complete already has the product data layer that UCP checkout reads.

Where UCP overlaps with ACP

ACP is the protocol UCP overlaps with most. Both let a shopper finish a purchase inside an AI surface: ACP was built around ChatGPT, and UCP's reference implementation runs in Google AI Mode and Gemini. The practical choice between them follows where your buyers already ask shopping questions.

Promptwatch's prompt tracking runs your buyer-intent prompts across ChatGPT, Google AI Mode and other AI platforms on a fixed schedule, which shows where your brand already surfaces before you commit integration work to one checkout surface.

Side-by-side comparison table

FAQ

Can I support more than one protocol at once?

Yes, and for most merchants that's the more defensible default than picking one. Nothing about AP2, ACP, MCP or UCP structurally prevents combining them, and UCP already builds on two of the others: it uses MCP as a transport and AP2 as a checkout extension. A Shopify store's catalog and cart tools now run through UCP over MCP, and the merchant can still add AP2 payment authorization or an ACP integration for ChatGPT on top without those choices conflicting.

Which protocol has the most retailer adoption today?

It depends on what you count. UCP has the most named retailers behind it, with Etsy, Target, Walmart and Wayfair as co-developers. AP2 has the deepest bench of payments partners. ACP had the most visible launch in ChatGPT, but in March 2026 OpenAI moved checkout from the chat to merchants' connected apps. Backer lists and live transaction volume are different signals, and none of the four protocols publishes purchase volume.

What happens if my chosen protocol loses out?

What needs reworking or retiring:

• Custom checkout logic built around that one protocol
• A single-vendor payment flow
• An API contract that assumed only that protocol mattered

What doesn't need touching:

• Structured product data
• A written liability policy
• Measurement infrastructure for agent-driven transactions

None of that second list is protocol-specific. ACP's own March 2026 scale-back is the concrete version of this: merchants who'd invested primarily in protocol-agnostic readiness kept that work; merchants whose whole plan was “integrate with ACP” had to start over on the integration piece.