What is the difference between UCP, ACP, and MCP?

UCP (Universal Commerce Protocol) by Google handles the full commerce lifecycle: product discovery, cart, order, and fulfillment. ACP (Agentic Commerce Protocol) by OpenAI and Stripe handles secure payment token exchange. MCP (Model Context Protocol) by Anthropic handles real-time data access from merchant systems to AI agents. They operate at different layers and work best together.

Do merchants need to choose between UCP, ACP, and MCP?

No. The three protocols operate at different layers of the agentic commerce stack and are designed to complement each other. UCP handles transactions, ACP handles payments, and MCP handles data access. The most capable merchant implementations use all three.

Which agentic commerce protocol should I implement first?

For most Shopify or DTC merchants, start with UCP for the broadest Google AI Mode reach. If you're already on Stripe, ACP is essentially automatic. Enterprise merchants with complex ERP systems should start with MCP to get data access right before layering on UCP for transactions.

Is MCP a commerce protocol?

MCP (Model Context Protocol) is not a checkout or payment protocol — it cannot complete a purchase on its own. It is a data access protocol that lets AI agents securely query merchant systems (inventory, pricing, customer data) in real time. It works alongside UCP and ACP in the full agentic commerce stack.

Who developed UCP, ACP, and MCP?

UCP was developed by Google in collaboration with Shopify and is endorsed by Visa, Mastercard, and major retailers. ACP was developed by OpenAI in partnership with Stripe. MCP was developed by Anthropic and has since been open-sourced and adopted broadly across the AI developer ecosystem.

UCP vs ACP vs MCP: The 2026 Agentic Commerce Protocol Comparison

The one-sentence answer: UCP (Google) runs the cart and order. ACP (OpenAI + Stripe) executes the payment. MCP (Anthropic) feeds real-time data to the agent. In the best 2026 implementations, all three run together. See also: How to Identify AI Agent Traffic in Google Analytics 4. See also: Cookieless Affiliate Tracking. See also: WebMCP Is Here. See also: 2026 UCP Compliance and Risk Checklist for Merchants.

Why This Comparison Matters Now

Three major AI labs have each released a protocol for agentic commerce in 2026. Merchants, developers, and enterprise architects are being asked to choose — or to explain to leadership why they might need all three. This guide cuts through the noise with a definitive breakdown of what each protocol actually does, where it operates in the stack, and when you need it.

Protocol Overview at a Glance

Protocol	Full Name	Developed By	Layer	Primary Function	Status (May 2026)
UCP	Universal Commerce Protocol	Google + Shopify	Transaction	Product discovery, cart, order, fulfillment	✅ Live — US merchants
ACP	Agentic Commerce Protocol	OpenAI + Stripe	Payment	Secure payment token exchange between agents and processors	✅ Live — Etsy, select Shopify
MCP	Model Context Protocol	Anthropic	Data Access	Real-time data retrieval from merchant systems (inventory, pricing, CRM)	✅ Live — Claude Marketplace

UCP — Universal Commerce Protocol

Developed by: Google, co-designed with Shopify. Endorsed by Visa, Mastercard, American Express, Stripe, Best Buy, Target, Walmart, Wayfair, and others.

What it does: UCP is an open REST standard that defines how AI agents interact with merchant storefronts end-to-end. It covers the entire commerce lifecycle: product catalog discovery, real-time inventory queries, cart creation, price negotiation, checkout initiation, and fulfillment status. It operates via the /.well-known/ucp discovery endpoint that merchants publish on their domain.

Who uses it: Any merchant wanting to sell through Google AI Mode, or any AI agent that follows the UCP standard for cross-platform shopping. Merchants implement UCP by publishing a structured endpoint and ensuring their product feed meets Google Merchant Center requirements.

Strengths: Broadest merchant adoption, backed by Google’s distribution, open standard with no vendor lock-in on the merchant side, works across all UCP-compliant agents (not just Google’s).

Limitations: Currently US-only for live checkout. Does not handle the payment layer directly — relies on processors like Google Pay. Requires Google Merchant Center approval before go-live.

ACP — Agentic Commerce Protocol

Developed by: OpenAI, built with Stripe as the payment infrastructure layer.

What it does: ACP focuses specifically on the payment execution step — the moment an AI agent needs to move money on behalf of a user. It defines how agents hold, present, and use Shared Payment Tokens: cryptographic credentials that authorize an agent to charge a stored payment method without exposing raw card data. ACP ensures the payment is authenticated, the merchant is legitimate, and the transaction is reversible if the agent made an error.

Who uses it: OpenAI’s ChatGPT Shopping uses ACP for its Etsy integration. Merchants who want their Shopify stores accessible via ChatGPT’s Instant Checkout flow need ACP-compatible payment handling (Stripe makes this automatic for Stripe-powered merchants).

Strengths: Purpose-built for agent payment security. Shared Payment Tokens prevent raw card data exposure to agents. Stripe’s network means instant compatibility for millions of merchants already on Stripe.

Limitations: OpenAI pivoted away from a centralized “Instant Checkout” model in March 2026, shifting toward a merchant-app architecture. ACP’s future roadmap is less certain than UCP’s. Does not handle product discovery or order management — payment only.

MCP — Model Context Protocol

Developed by: Anthropic. Open-sourced and now adopted across the broader AI developer ecosystem beyond Claude.

What it does: MCP is the data access layer of the agentic stack. It defines how an AI agent (Claude or any MCP-compatible agent) connects to external systems — your ERP, inventory database, pricing engine, or CRM — to pull real-time context before making commerce decisions. MCP is not a checkout protocol. It’s a secure, permissioned API bridge: the merchant runs an MCP server that defines exactly what the agent can query, read, and write.

Who uses it: Enterprise merchants who need Claude to access their proprietary systems (SAP, Oracle, Coupa) without building custom integrations. Developers building commerce agents that need real-time inventory or personalization data. See how Claude Marketplace uses MCP for agentic commerce.

Strengths: Protocol-agnostic — an MCP server can feed data to agents running on UCP, ACP, or any other protocol. Merchant controls the data surface entirely. Strong GDPR alignment (data stays in merchant’s infrastructure). Growing developer ecosystem with tools from GitHub, Cloudflare, and others.

Limitations: Not a checkout or payment protocol — cannot complete a purchase on its own. Requires merchant to run and maintain an MCP server. No equivalent to Google’s broad distribution advantage.

How They Fit Together: The Full-Stack Architecture

The three protocols are not competitors — they operate at different layers of the same transaction. Here is how a complete agentic commerce transaction flows through all three:

Agent receives user intent (“Buy me new running shoes under $120, size 10, same brand as last time”)
MCP layer: Agent queries the merchant’s MCP server for real-time inventory, the user’s purchase history, and current pricing rules
UCP layer: Agent uses the merchant’s /.well-known/ucp endpoint to confirm product availability, create a cart, and initiate checkout
ACP layer: Agent presents a Shared Payment Token to Stripe; Stripe authorizes the charge against the user’s stored payment method
UCP layer: Order confirmation and fulfillment status flow back through UCP to the agent and user

A merchant fully implemented on all three layers is reachable by any AI agent — Google’s, Anthropic’s, OpenAI’s, or a third-party agent built on any of these standards.

Decision Guide: When to Prioritize Which Protocol

Your Situation	Start With	Then Add
Shopify merchant wanting Google AI Mode visibility	UCP	MCP for personalization
Etsy seller or Stripe-powered DTC brand	ACP	UCP for broader agent reach
Enterprise with SAP/Oracle ERP, B2B focus	MCP	UCP for transaction layer
EU merchant with strict GDPR requirements	MCP (data residency control)	UCP with GDPR-compliant processors
Regulated industry (healthcare, financial services)	MCP (audit trail control)	UCP after compliance clearance
DTC brand wanting maximum agent reach	UCP + ACP simultaneously	MCP for personalization layer

The Protocol War Is Not a Zero-Sum Game

Early coverage framed UCP vs ACP vs MCP as a standards war with a single winner. That framing is wrong. These protocols solve different problems at different layers of the same stack. Google has no incentive to build a payment token protocol — Stripe does that better. Anthropic has no incentive to replicate Google’s distribution network — UCP does that better. The interoperability is intentional.

The merchants who will win in agentic commerce are not those who pick the right protocol — they’re the ones who implement all three and become reachable by every agent in the ecosystem. The technical cost of implementing all three is lower than most merchants expect: UCP requires the discovery endpoint and product feed; ACP is automatic for Stripe merchants; MCP requires running a server but the tooling has matured significantly in early 2026.