ACP vs UCP: The Two Protocols That Will Decide How AI Agents Buy and Sell in 2026

September 2025: The Starting Gun

On September 29, 2025, Stripe and OpenAI did something that had never been done before: they let an AI agent complete a real purchase on behalf of a human — inside a chat interface, with no browser, no cart page, and no checkout form.

The feature was called Instant Checkout. The protocol underneath it was the Agentic Commerce Protocol (ACP). Within weeks, Etsy sellers were processing orders that originated entirely inside ChatGPT. Shopify announced that over one million of its merchants — including Glossier, Vuori, Spanx, and SKIMS — would follow.

Three months later, Google CEO Sundar Pichai took the stage at the National Retail Federation conference in New York and unveiled a competing vision: the Universal Commerce Protocol (UCP), co-developed with Shopify, and backed by Walmart, Target, Wayfair, Etsy, Mastercard, Visa, and 20+ other partners.

Two protocols. Two ecosystems. One question every ecommerce operator is now asking: which one do I need?

The answer, as we will see, is probably both.

The Agentic Commerce Protocol (ACP): How It Works

ACP was co-developed by OpenAI and Stripe and released as an open standard under the Apache 2.0 license. The specification is published at agenticcommerce.dev and the source code lives on GitHub under the agentic-commerce-protocol organization.

The protocol defines a four-party flow between a buyer, an AI agent, a business, and a payment provider:

The ACP Four-Party Flow

1. Buyer asks a shopping question in an AI surface (e.g. "best running shoes under $100"). The AI agent shows relevant products — ranked by relevance, not ads.
2. AI Agent displays product details, collects the buyer's payment credentials, and sends a checkout request to the merchant.
3. Business receives the checkout request with a secure payment token. The merchant validates the order, calculates tax and shipping, runs fraud checks on their own stack, and accepts or declines.
4. Payment Provider (Stripe, PayPal, or any compatible processor) facilitates the actual charge. The merchant remains the merchant of record.

The critical design principle: the merchant retains full control. The merchant decides what products appear, how the brand is presented, which orders to accept, and which payment processor to use. The AI agent handles discovery and presentation; the merchant handles everything else.

What ACP Supports

• Physical goods, digital goods, subscriptions, and asynchronous purchases
• Custom checkout configurations including in-store pickup and dynamic pricing
• Any compatible payment provider — not just Stripe
• Shared Payment Tokens (SPTs) that let merchants accept agentic payments without switching payment processors

Who Has Adopted ACP

The adoption timeline moved fast:

• September 29, 2025 — ACP launches with Instant Checkout in ChatGPT. Etsy sellers go live immediately.
• October 14, 2025 — Salesforce announces ACP integration, building Instant Checkout into its commerce cloud in collaboration with Stripe and OpenAI.
• October 28, 2025 — PayPal joins the ecosystem, bringing its global merchant network and enabling PayPal wallet checkout inside ChatGPT.
• December 11, 2025 — Stripe launches the Agentic Commerce Suite, a low-code implementation layer that bundles product discovery, checkout management, and payment processing into a single integration. Early adopters include URBN (Anthropologie, Free People, Urban Outfitters), Coach, Kate Spade, Revolve, Ashley Furniture, and Halara.

The Suite also brought in ecommerce platforms: Wix, WooCommerce, BigCommerce, Squarespace, and commercetools. And omnichannel partners: Akeneo, Cymbio, Logicbroker, Mirakl, Pipe17, and Rithum.

Stripe's pitch to merchants was simple: if you already process payments with Stripe, you can enable agentic commerce with approximately one line of code. If you use a different processor, you can still participate via Stripe's Shared Payment Token API or the Delegated Payments Spec.

ACP's Partners Beyond Checkout

Stripe also named Microsoft Copilot, Anthropic, Perplexity, Vercel, Lovable, Replit, Bolt, and Manus as early partners testing agentic commerce solutions — signaling that ACP is not limited to ChatGPT.

The Universal Commerce Protocol (UCP): How It Works

UCP was co-developed by Google and Shopify, with co-development contributions from Etsy, Wayfair, Target, and Walmart. Google CEO Sundar Pichai unveiled it at the National Retail Federation conference on January 11, 2026.

Shopify CEO Tobi Lütke put it directly: “Shopify is building the foundation for agentic commerce. Universal Commerce Protocol, which we co-developed with Google, is now live. UCP will make it faster for agents and retailers to integrate. It’s open by default, so platforms and agents can use UCP to start transacting.”

Where ACP focuses on the checkout transaction, UCP covers the entire shopping lifecycle:

UCP's Full-Lifecycle Architecture

Shopify's engineering team built UCP using a layered protocol model inspired by TCP/IP. Three layers handle different concerns:

1. Shopping Service Layer — Core transaction primitives: checkout sessions, line items, totals, messages, and status.
2. Capabilities Layer — Major functional areas (Checkout, Orders, Catalog) with independent versioning. Merchants and agents declare what they support; the protocol negotiates the intersection.
3. Extensions Layer — Domain-specific schemas for discounts, fulfillment rules, loyalty programs, and subscriptions. Organizations own namespaces tied to their domains — no central registry needed.

How Capability Discovery Works

Every UCP-enabled merchant publishes a manifest at /.well-known/ucp — a JSON document declaring what the merchant supports. AI agents query this manifest to understand what they can do: browse products, apply discounts, initiate checkout, track orders, process returns.

This is dynamic negotiation. Instead of rigid integrations built in advance, agents and merchants discover each other's capabilities at transaction time. If a merchant adds loyalty support next month, agents automatically detect it — no integration update required.

The Checkout State Machine

UCP implements a three-state checkout flow:

• Incomplete — Required information is missing. The agent attempts to resolve it via API calls (address, shipping method, payment).
• Requires Escalation — The agent cannot resolve a requirement programmatically. The buyer is handed off to a continue_url where they complete the step in the merchant's own interface.
• Ready for Complete — All data collected. The agent finalizes the purchase programmatically.

This escalation model is key: UCP does not assume the AI agent can handle everything. When it hits a complex scenario (age verification, custom engraving, appointment scheduling), it gracefully hands control to the merchant's UI.

Transport Flexibility

UCP supports three transport layers:

• REST APIs — Standard HTTP for traditional integrations
• Agent-to-Agent (A2A) — Google's protocol for agent-to-agent communication
• Model Context Protocol (MCP) — Anthropic's standard for AI-to-application connectivity

You can swap transport mechanisms without changing business logic. A merchant that starts with REST can add MCP support later without rewriting their commerce layer.

Open Payments

Unlike ACP's initial Stripe-centric design, UCP's payment architecture is processor-agnostic from the start. Payment handlers are negotiated — not prescribed. Agents declare what credentials they support (saved cards, wallets, BNPL). Merchants advertise which handlers they accept based on cart, location, and amount. Shop Pay is built in for Shopify merchants, but Google Pay, PayPal, or any other processor can plug in through the same mechanism.

Who Has Adopted UCP

The partner list is broad:

• Co-developers: Google, Shopify, Etsy, Wayfair, Target, Walmart
• Endorsing partners (20+): Adyen, American Express, Best Buy, Flipkart, Home Depot, Macy's, Mastercard, Stripe, Visa, Zalando, and more
• Live checkout (Feb 2026): Etsy and Wayfair in Google AI Mode and Gemini app for U.S. shoppers
• Coming soon: Shopify, Target, and Walmart integrations

The specification is published at ucp.dev with a public GitHub repository at universal-commerce-protocol/ucp featuring Python SDK samples, reference implementations, and active community discussions.

ACP vs UCP: The Real Differences

The two protocols are frequently positioned as competitors. They are not — at least, not yet. They occupy different layers of the agentic commerce stack.

Dimension	ACP (OpenAI + Stripe)	UCP (Google + Shopify)
Launched	September 29, 2025	January 11, 2026
Primary scope	Checkout and payment	Full shopping lifecycle (discovery → checkout → post-purchase)
Primary AI surface	ChatGPT Instant Checkout	Google AI Mode, Gemini, any agent
Payment default	Stripe (others via SPT API)	Processor-agnostic (negotiated)
Product discovery	Organic, unsponsored ranking	Merchant catalog + capability negotiation
Post-purchase	Limited (merchant handles)	Full (tracking, returns, support)
Transport layers	ACP-specific API	REST, A2A, MCP
License	Apache 2.0	Open source (ucp.dev)
Merchant integration	~1 line of code (Stripe merchants)	/.well-known/ucp manifest + capabilities
Key partners	Salesforce, PayPal, Shopify, Perplexity	Walmart, Target, Wayfair, Mastercard, Visa

Notice something? Stripe and Shopify appear on both sides. Stripe endorsed UCP. Shopify merchants are joining ACP through Instant Checkout. This is not a zero-sum contest — it is a layering.

How the Protocols Fit Together

Think of it as a stack:

• MCP (Anthropic) — The data and tool layer. How AI agents connect to inventory systems, order databases, and operational tools.
• UCP (Google + Shopify) — The commerce lifecycle layer. How agents discover products, negotiate capabilities, handle checkout, and manage post-purchase.
• ACP (OpenAI + Stripe) — The checkout execution layer. How agents complete a specific purchase transaction with payment processing.

A practical scenario: a customer asks an AI agent to find a winter jacket. The agent uses MCP to connect to a merchant's inventory system and check real-time stock levels. It uses UCP to discover the merchant's product catalog, compare options, and negotiate available shipping methods. When the customer says "buy this one," the agent uses ACP to process the payment through the merchant's Stripe account and complete the order.

Three protocols, three layers, one transaction.

The Merchant's Dilemma: What Should You Actually Do?

If you are running an ecommerce operation in 2026, here is the practical reality:

If You Are on Shopify

UCP support is being built into the platform natively. You will get agent-compatible product discovery, checkout, and order management without writing protocol-level code. Shopify is also integrating with ACP — your products will be discoverable in ChatGPT Instant Checkout through the existing Shopify + Stripe integration.

Action: Ensure your product catalog data is clean (titles, descriptions, images, variant attributes) because AI agents will use it to decide which products to show users. Poor catalog data means your products get ranked lower or excluded entirely.

If You Are on Stripe

ACP integration is minimal — potentially one line of code to enable agentic payments. The Agentic Commerce Suite handles product feeds, checkout sessions, and fraud protection. You can join the waitlist through Stripe Dashboard.

Action: Enable agentic payments in Stripe. If you are not on Shopify, consider implementing a UCP manifest (/.well-known/ucp) to make your store discoverable by Google AI Mode.

If You Are on WooCommerce, BigCommerce, or Other Platforms

Both protocols are accessible: Stripe's Agentic Commerce Suite lists WooCommerce, BigCommerce, and Squarespace as supported platforms. UCP is open source and platform-agnostic — any merchant can implement a /.well-known/ucp manifest.

Action: Start with the protocol that matches your existing payment stack. If you use Stripe, start with ACP. If your primary discovery channel is Google, prioritize UCP.

If You Are a Marketplace Seller (Etsy, Amazon, Walmart)

Etsy is already live on both ACP (Instant Checkout in ChatGPT) and UCP (checkout in Google AI Mode). Walmart is a UCP co-developer. Amazon has not joined either protocol.

Action: If you sell on Etsy, your products are already discoverable by AI agents. If you sell on Amazon, you are currently locked into Amazon's proprietary ecosystem (Rufus). Diversifying to a direct-to-consumer channel with ACP/UCP support is increasingly strategic.

The Criticisms: What Could Go Wrong

These protocols are not without problems. Industry analysts and merchants have raised several concerns:

The Double-Stack Problem

If ACP and UCP do not converge, merchants face the burden of maintaining two separate integrations: one optimized for ChatGPT's checkout flow, another for Google's full-lifecycle model. For large brands with engineering teams, this is manageable. For a 10-person DTC brand, it is another tax on already-stretched resources.

Platform Power Consolidation

Both protocols shift discovery power upstream. When an AI agent decides which products to show a shopper, the merchant loses control of the browsing experience. Product rankings in AI surfaces are opaque — there is no "SEO" for agentic commerce yet. The platforms that control discovery (Google, OpenAI) gain leverage, while merchants carry the operational complexity.

Stripe Lock-In Concerns

ACP's payment flow is optimized for Stripe. While the Shared Payment Token API and Delegated Payments Spec allow other processors, the easiest path runs through Stripe. Merchants using legacy payment gateways may need to build bridges or add Stripe as a secondary processor specifically for agentic payments.

The Amazon Question

Amazon — the largest ecommerce platform in the U.S. — has not joined either protocol. With its Rufus AI agent and closed marketplace, Amazon has little incentive to support a neutral standard that helps independent merchants compete. If Amazon builds its own proprietary agent commerce interface, the market fragments into three ecosystems instead of converging on two.

Brand Differentiation Erosion

When products are surfaced through AI agents, the shopping experience becomes transactional. There is no brand storytelling, no carefully designed product page, no lifestyle imagery. UCP's capability negotiation helps (merchants can declare support for rich media, loyalty programs, and custom experiences), but the default agent interaction is utilitarian: here are three options, which one do you want?

What Happens Next

We are in the TCP/IP moment of agentic commerce. Just as the internet needed HTTP to standardize how browsers talk to servers, AI commerce needs ACP and UCP to standardize how agents talk to merchants.

Several things to watch:

• Protocol convergence: Stripe is already a UCP endorser. Shopify merchants are on ACP. The overlap will grow. Whether the protocols formally merge, interoperate via adapters, or remain separate standards with dual-implementation is the defining question of the next 12 months.
• Amazon's response: If Amazon launches its own protocol, merchants face a three-way split. If Amazon adopts UCP or ACP, we get consolidation. Neither outcome is predictable.
• Merchant tooling: The Agentic Commerce Suite and Shopify's native UCP support lower the integration bar. But for merchants not on these platforms, implementation still requires engineering work. Third-party middleware (Pipe17, Rithum, commercetools) will likely build cross-protocol adapters.
• Regulation: AI agents processing payments on behalf of consumers will attract regulatory attention. Who is liable when an AI agent buys the wrong product? Who handles chargebacks? Both protocols designate the merchant as merchant of record, but edge cases will test this framework.
• Consumer trust: Today, most consumers do not trust AI agents to make purchases on their behalf. Building that trust requires flawless execution — one bad experience (wrong product, unauthorized charge, missing order) sets the entire category back.

The Bottom Line for Ecommerce Operators

The agentic commerce era is not coming — it is here. Etsy sellers are processing orders from ChatGPT. Wayfair shoppers are checking out inside Google AI Mode. The protocols are live, the infrastructure is real, and the consumer behavior is shifting.

For operators, the playbook is straightforward:

1. Clean your catalog data. AI agents rank products based on structured data quality. Titles, descriptions, images, variant attributes, and inventory accuracy are now competitive advantages.
2. Enable ACP if you use Stripe. The integration is trivial and gives you access to ChatGPT's user base.
3. Prepare for UCP. If you are on Shopify, it is coming natively. If not, start familiarizing yourself with the specification at ucp.dev.
4. Maintain your MCP layer. Both ACP and UCP benefit from clean, real-time data connections. An MCP server that exposes your inventory, orders, and product data makes every protocol integration stronger.
5. Do not pick sides. ACP and UCP are complementary layers, not competing standards. The merchants who win will be the ones present on both — just as the best retailers today sell on both Google Shopping and Amazon, not one or the other.

The question is no longer whether AI agents will buy products. They already are. The question is whether your store is ready to sell to them.

Nventory's ACP and UCP Support: Early Access

Nventory is actively building beta support for both ACP and UCP. Merchants on Nventory will be able to expose their product catalogs, inventory, and checkout flows to AI agents through both protocols — without building separate integrations for each.

Combined with Nventory's existing production MCP server, this means your store data is accessible to every major AI surface: Claude and ChatGPT via MCP, ChatGPT Instant Checkout via ACP, and Google AI Mode via UCP.

If you are interested in early access to ACP or UCP integration on Nventory, reach out to our team. We are working with merchants to test agentic checkout flows and would welcome your feedback as we shape the implementation.

Related Reading

• The AI-Native OMS: How MCP, OpenClaw, and Conversational AI Are Replacing the Dashboard — Deep dive into Nventory's MCP server, OpenClaw compatibility, and conversational AI Suite.
• OpenClaw for Ecommerce: How the Fastest-Growing AI Agent Is Reshaping Operations — How OpenClaw connects messaging platforms to commerce systems via MCP.