What Are ACP, UCP, and MCP? The Agentic Commerce Protocols Explained

Agentic commerce is the shift from a human clicking through a storefront to an AI agent autonomously discovering products, building a cart, and completing a purchase. Three protocols define how that works: MCP handles the underlying tool connectivity, ACP covers checkout and payment, and UCP standardises the full commerce journey from discovery to post-purchase. Each has a distinct scope, a distinct backer, and a distinct moment in the transaction flow.

What Is MCP?

MCP (Model Context Protocol) is the foundation layer. Created by Anthropic and now governed by the Linux Foundation, it provides a general-purpose protocol for connecting AI models to external tools and data sources, including commerce APIs. It is not commerce-specific.

Think of MCP as the universal adapter that lets an AI agent query a product catalogue, check inventory, pull order history, or trigger an action in any backend system. MCP defines how to expose resources (data), tools (callable functions), and prompts (predefined instructions) through a standardised interface. An AI agent running on any platform can connect to an MCP server and immediately understand what capabilities are available, without bespoke integration work.

For merchants, this means an MCP server in front of your product catalogue allows any compatible AI agent, whether from Salesforce, Adobe, or a custom enterprise deployment, to query your inventory and product data without a custom API contract for each one.

What Is ACP?

ACP (Agentic Commerce Protocol) sits on top of the MCP foundation and focuses narrowly on checkout and payment. Built by Stripe and OpenAI, it handles commerce flows inside ChatGPT and went live in production in late 2025, with PayPal and Worldpay as initial payment partners.

ACP provides two primary interfaces. The Checkout API handles cart management: an AI agent can add items, apply promotions, and submit an order through a standardised REST interface. The Delegate Payment API handles secure payment tokenisation through payment service providers like Stripe, so an AI agent can authorise a purchase without the user exposing card details to the agent directly.

The minimal surface area is intentional. ACP does not try to solve product discovery or the post-purchase experience. It solves the moment the cart is ready and the user wants to pay, nothing more. That constraint is what made it possible to ship in production quickly and integrate with existing PSP infrastructure that merchants already use.

What Is UCP?

UCP (Universal Commerce Protocol) is broader. Launched by Google with Shopify and Walmart at NRF 2026, it covers the full commerce journey: product discovery, comparison, cart management, checkout, and post-purchase flows through a single standardised abstraction layer.

The problem UCP is solving is the N-times-N integration problem. Without a shared protocol, every retailer must build separate integrations for Google AI Mode, Gemini, ChatGPT, Perplexity, and every other AI surface that wants to sell their products. UCP collapses this to a single connection. Google describes UCP as compatible with existing protocols, including its own Agent Payment Protocol, meaning ACP and UCP can coexist rather than compete.

UCP is the right protocol when the AI surface needs to handle the entire purchase arc, not just the final payment step. A user asking Google AI Mode to "find me a winter running jacket under $150 with free returns" needs the AI to query, compare, and present options before a cart even exists. That discovery-to-checkout scope is where UCP operates.

How the Three Protocols Differ

The clearest way to separate the three is by scope and use case:

Protocol	Backed By	Scope	Live Since	Best For
MCP	Anthropic / Linux Foundation	Tool connectivity, data access	2023	Enterprise AI, catalogue integration
ACP	Stripe + OpenAI	Checkout + payment only	Late 2025	ChatGPT transactions
UCP	Google + Shopify + Walmart	Full commerce journey	2026	Google AI Mode, Gemini

MCP is commerce-agnostic. It was not designed for retail but is being adopted by enterprise AI tools for catalogue connectivity because it already handles the authentication and tool-exposure patterns they need. ACP and UCP are purpose-built for commerce but at different points in the funnel.

How They Work Together in One Transaction

A single AI-driven purchase does not pick one protocol and ignore the others. The protocols are designed to layer. A realistic ChatGPT checkout scenario in mid-2026 might look like this:

MCP handles tool connectivity. The AI agent authenticates with the merchant's backend, queries product data, and retrieves real-time inventory through an MCP server.
UCP handles product discovery. If the merchant has implemented UCP, the agent can search the catalogue with a standardised interface, compare options, and present results in a structured format.
ACP handles checkout. Once the user selects a product, ACP's Checkout API builds the cart and applies shipping and promotional logic.
ACP handles payment. The Delegate Payment API tokenises the user's saved payment method through a PSP like Stripe, without transmitting raw card data to the agent.
Order is confirmed. The merchant's backend processes the order through existing infrastructure, with no storefront visit required.

Most retailers will not need all three simultaneously on day one. A Shopify merchant already on Google's UCP programme is covered for discovery and checkout on Google surfaces. Adding an MCP server for enterprise catalogue access is an independent decision based on whether B2B or enterprise AI channels matter to their business.

Channel Dictates Protocol

The practical rule for merchants deciding where to invest is simple: your channel dictates your protocol.

ChatGPT and OpenAI agents use ACP for commerce transactions. If ChatGPT is a significant traffic or conversion channel for your category, ACP is where to focus first. PayPal and Worldpay are the initial payment partners, with Stripe as the underlying infrastructure.
Google AI Mode and Gemini use UCP. Google's AI surfaces are integrated with Shopify and Walmart natively. Google AI Mode launched in the US and is expected to expand to additional markets, which makes UCP integration via Shopify or direct Google Merchant Centre adoption worth prioritising for merchants already active on Google Shopping.
Enterprise AI tools rely on MCP. Salesforce Agentforce, Adobe Experience Cloud, and SAP's AI capabilities increasingly use MCP for catalogue integration. If B2B wholesale, enterprise procurement, or large-account sales are part of your model, an MCP server in front of your product and inventory data is the relevant investment.

Most serious retailers will end up supporting at least two of these protocols. The good news is they are designed to coexist. Implementing UCP via a Shopify integration does not conflict with separately building an MCP server for enterprise access, and ACP's checkout layer works on top of both.

SEO and AI Commerce Implications

These protocols change where visibility and conversion happen. When an AI agent handles product discovery and checkout, the traditional funnel, search to click to PDP to cart, is compressed or bypassed entirely. SEO for agentic commerce means making your product data legible to protocol-based retrieval, not just crawlable to a traditional bot. For a broader look at how AI agents discover and select businesses across all verticals, see our guide to agentic SEO.

Example: How AI Agents Retrieve Product Data

Without protocol support (agent fails to retrieve)

An AI agent queries your product catalogue through a generic URL crawl. Prices are rendered client-side via JavaScript and are not accessible to the agent. Inventory status is not exposed. The agent returns a competitor product with available data instead.

With protocol support (agent retrieves and sells)

The agent calls your MCP server, retrieves current price, stock status, and variant data in structured JSON. It passes this to UCP's discovery layer. The user sees your product with accurate details and completes checkout via ACP in the same session.

Expose structured product data via protocol-compatible endpoints. Structured Data (Product schema with price, availability, and identifier) is the minimum. For serious agentic commerce volume, implement an MCP server or register with Shopify's UCP feed. Agents that cannot retrieve accurate product data reliably will surface competitors who can.
Treat your product feed as an SEO asset. UCP and Google's Shopping Graph rely on product feed quality for accurate retrieval and presentation. Missing GTINs, inconsistent pricing, and stale inventory data are likely to reduce your visibility in Google AI Mode, much as low-quality on-page content reduces organic visibility, though Google has not published specific feed quality ranking weights for AI surfaces.
Prioritise the channel where your category transacts. Intent signals are likely to differ by surface, though published data on user behaviour across ACP and UCP channels is still limited at this stage of rollout. Check your own analytics for referral patterns from ChatGPT and Google AI surfaces, and allocate protocol investment based on where your actual buyers already appear, not on which protocol launched first.
Write product content for agent comprehension, not just human readers. AI agents parsing product descriptions look for entity-attribute-value patterns. "Waterproof, weight 280g, compatible with ISO 9001-certified harnesses" is more legible to an agent than "the perfect jacket for any adventure." Clear attributes in your content reduce the chance of an agent misrepresenting your product.
Monitor referral traffic source shifts. As agentic commerce grows, direct and referral attribution will change. Orders arriving via ACP from ChatGPT may not carry standard UTM parameters. Set up server-side order attribution that tracks agent-initiated purchases separately. Work with our AI SEO team to build measurement frameworks that capture this new traffic pattern.

Frequently Asked Questions

Do ACP, UCP, and MCP compete with each other?

No. The three protocols are designed to work at different layers of the same transaction. MCP provides general-purpose tool connectivity, ACP handles checkout and payment, and UCP covers the full commerce journey. A single purchase can use all three sequentially. Google has explicitly described UCP as compatible with existing protocols including its own Agent Payment Protocol, and ACP's checkout layer sits on top of any discovery mechanism.

Which protocol should a small to mid-size retailer implement first?

Start with the channel that already sends you traffic and buyers. If you are on Shopify and Google Shopping is a meaningful channel, UCP via Shopify is the lowest-friction path since Shopify is a founding UCP partner. If ChatGPT referrals are visible in your analytics, explore ACP readiness through your PSP. MCP is most relevant for B2B, enterprise, or multi-platform catalogue exposure scenarios.

Is MCP only for enterprise businesses?

MCP is general-purpose and not limited by business size. Smaller merchants can implement an MCP server to expose their product catalogue to any compatible AI agent. In practice, MCP adoption is currently strongest in enterprise contexts because Salesforce, Adobe, and SAP have adopted it for their AI tools. But as the open standard matures under Linux Foundation governance, more SMB-friendly implementations are emerging.

How does the Delegate Payment API in ACP protect user payment data?

The Delegate Payment API uses tokenisation through a payment service provider (initially Stripe, PayPal, and Worldpay). The AI agent never sees raw card data. Instead, the PSP holds the payment instrument and returns a secure token that the agent uses to authorise the transaction. This mirrors how guest checkout with a saved card works in a conventional storefront, but the flow is initiated by the agent rather than the user clicking.

Will these protocols replace traditional e-commerce storefronts?

Not immediately and probably not fully. These protocols add an agentic channel alongside the existing storefront. Most purchases, particularly high-consideration or first-time category purchases, will continue to involve a human visiting a product page and making a deliberate decision. Agentic commerce is most likely to capture replenishment purchases, commodity products, and scenarios where the user has strong brand or product preferences that an agent can represent reliably.

How should merchants track sales that come through ACP or UCP?

Agent-initiated purchases often arrive without standard UTM parameters because the agent does not navigate through a URL chain in the way a human browser session does. Merchants should implement server-side order attribution that flags orders initiated through ACP or UCP endpoints separately from direct and organic traffic. The specifics of how each protocol surfaces channel data in reporting will depend on your PSP and platform integrations, and are evolving as both ACP and UCP move beyond early rollout.

Prepare Your Site for Agentic Commerce

Whether you need to assess which protocol is the right priority for your channel mix, implement structured data for agent-readable product content, or build an attribution framework for AI-driven orders, our team can help you navigate the shift from keyword search to agentic transactions.