MCP, ACP, UCP, A2A: The Four Protocols Powering AI Commerce

Four protocols now define how AI agents interact with commerce. Each was created by a different coalition of technology companies to solve a different problem. Together, they form the emerging infrastructure layer for agentic commerce. Separately, they represent a fragmentation challenge that every merchant and developer must navigate.

This article provides a comprehensive technical and strategic analysis of each protocol: what it does, who built it, how it works, where it stands in adoption, and how it relates to the others.

MCP: The Connectivity Standard

Model Context Protocol (MCP) is an open standard released by Anthropic in November 2024. Its purpose is straightforward: provide a universal way for AI assistants to connect to external data sources and tools.

Before MCP, every AI integration was bespoke. If a developer wanted an AI assistant to access a database, query an API, or interact with a service, they built a custom connector. MCP standardizes that connection layer, providing a common protocol so that any MCP-compatible AI system can interact with any MCP-compatible service.

The adoption trajectory has been remarkable. OpenAI adopted MCP in March 2025. Google, Microsoft, and Shopify followed. In summer 2025, Shopify activated a default MCP endpoint at /api/mcp on every store on its platform, making millions of merchants instantly accessible to MCP-compatible AI agents. The November 2025 spec update (version 2025-11-25) marked the one-year anniversary with refinements based on real-world implementation feedback.

PayPal's head of AI publicly stated that MCP was the moment he realized agentic e-commerce was real. That sentiment captures the protocol's significance: MCP did not invent AI-to-service connectivity, but it standardized it in a way that catalyzed the entire ecosystem.

Technical characteristics: MCP defines a client-server architecture where AI applications (clients) connect to data sources and tools (servers) through a standardized interface. It supports multiple transport mechanisms and provides a schema for describing available tools, their parameters, and their outputs.

In the commerce context, MCP enables AI agents to query product catalogs, check inventory, read shipping policies, and access merchant data in a structured, predictable format. It is the foundation layer that other protocols build upon or complement.

Current status: De facto standard for AI-to-service connectivity. The broadest adoption of any protocol in this space, with support from every major AI platform.

ACP: The Payment Layer

Agentic Commerce Protocol (ACP) was announced by OpenAI and Stripe in September 2025. Licensed under Apache 2.0, it is an open standard designed specifically for programmatic commerce between AI agents and businesses.

ACP addresses a fundamental problem: how does an AI agent pay for something on behalf of a user without exposing the user's payment credentials? Traditional checkout flows assume a human is present to enter card details, authenticate with a bank, and confirm the purchase. AI agents need a different mechanism.

The core innovation is Shared Payment Tokens (SPTs). These are tokenized payment instruments that an AI agent can use to initiate a transaction on the user's behalf. SPTs are scoped to specific sellers, bounded by time limits and maximum amounts, and never expose the underlying payment credentials. A consumer authorizes an SPT with defined constraints, and the agent operates within those boundaries.

OpenAI and Stripe serve as Founding Maintainers of the protocol, which is publicly available on GitHub at github.com/agentic-commerce-protocol. Early adopters include URBN (parent company of Anthropologie, Free People, and Urban Outfitters), Etsy, Ashley Furniture, Coach, Kate Spade, and Revolve.

In production, ACP powers Instant Checkout within ChatGPT. When a ChatGPT user decides to purchase something, ACP handles the transaction flow: the agent presents the offer, the user authorizes payment through an SPT, and Stripe processes the transaction. Etsy sellers were the first live merchants; Shopify merchants are in the integration pipeline.

Technical characteristics: ACP defines the message formats, authentication flows, and transaction lifecycle for agent-initiated commerce. It specifies how agents discover purchasable items, how payment authorization works, how transactions are confirmed, and how disputes are handled.

In the commerce context, ACP is the payment rail for agentic commerce. Where MCP connects agents to data, ACP connects agents to checkout. It solves the credential exposure problem that would otherwise prevent agents from completing purchases.

Current status: Live in ChatGPT Instant Checkout with Etsy merchants. Growing adoption among major retail brands. The payment-specific focus makes it complementary to, rather than competitive with, the other protocols.

UCP: The Shopping Journey Standard

Universal Commerce Protocol (UCP) was announced on January 29, 2025, co-developed by Google and Shopify. It is an open standard that covers the full shopping journey: product discovery, purchasing, and post-purchase interactions including returns, tracking, and customer support.

While MCP provides general AI-to-service connectivity and ACP handles payments, UCP is purpose-built for the complete commerce experience. It defines how an AI agent should discover products, present options to consumers, facilitate purchasing decisions, handle order tracking, and manage returns. This end-to-end scope is its distinguishing feature.

The endorsement list is extensive: more than twenty partners including Etsy, Wayfair, Target, Walmart, Adyen, American Express, Best Buy, Flipkart, Macy's, Mastercard, Stripe, The Home Depot, Visa, and Zalando. That coalition spans marketplaces, retailers, payment networks, and card brands, representing a significant portion of global commerce infrastructure.

Technical characteristics: UCP supports multiple transport mechanisms, including REST APIs, MCP, an Agent Payments Protocol (AP2), and A2A. This architectural flexibility means that a UCP-compatible service can be accessed through whichever transport mechanism an agent prefers. The protocol defines schemas for product data, offers, inventory, order management, fulfillment tracking, returns, and customer service interactions.

Consumer surfaces for UCP include AI Mode in Google Search and the Gemini app, where Google's AI agents use UCP to interact with merchant catalogs and facilitate purchases.

The protocol's website is ucp.dev, where the full specification and implementation guides are available.

In the commerce context, UCP provides the most comprehensive coverage of any single protocol. A fully UCP-compliant merchant exposes their entire shopping experience to AI agents in a standardized format, from the first product query through post-purchase support.

Current status: Strong backing from major retailers and payment networks. Deployed in Google's consumer AI products. The breadth of its scope makes it ambitious; the depth of its backing makes it credible.

A2A: The Agent Collaboration Layer

Agent2Agent Protocol (A2A) was announced by Google in April 2025. It addresses a different problem from the other three protocols: how do AI agents from different systems communicate and collaborate with each other?

The previous three protocols all define how an agent interacts with a service (a data source, a payment system, a merchant). A2A defines how agents interact with other agents. In a commerce context, this enables scenarios like a buyer's agent negotiating with a seller's agent, a logistics agent coordinating with a fulfillment agent, or a customer service agent collaborating with a returns processing agent.

A2A launched with more than fifty partners, including Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, PayPal, Salesforce, SAP, and ServiceNow. In June 2025, the protocol was contributed to the Linux Foundation, signaling a commitment to vendor-neutral governance. By July 2025, version 0.3 introduced gRPC support, and the consortium had grown to over 150 organizations.

Technical characteristics: A2A is built on HTTP, Server-Sent Events (SSE), and JSON-RPC. Key features include Agent Cards (JSON-formatted descriptions of an agent's capabilities), task lifecycle management, context sharing between agents, and UX negotiation (how agents coordinate the user experience when multiple agents are involved in a single workflow).

In the commerce context, A2A enables multi-agent workflows. A personal shopping agent could discover products via MCP, negotiate price through A2A with a merchant's agent, complete payment via ACP, and track delivery through UCP. Each protocol handles its layer; A2A handles the inter-agent coordination.

Current status: Broad industry support with over 150 organizations. Linux Foundation governance provides neutrality. Still maturing in terms of real-world commerce deployments, but the foundational infrastructure is in place.

Protocol Comparison

The Fragmentation Problem

For developers building AI agents, protocol diversity is manageable. For merchants trying to be discoverable and transactable across all AI platforms, it is a genuine operational burden.

Consider the practical reality. A ChatGPT user shopping through Instant Checkout triggers ACP for payment. A Google AI Mode user browsing products uses UCP for discovery and purchasing. A developer building a Claude-based shopping assistant connects via MCP. An enterprise deploying multi-agent procurement workflows uses A2A for coordination.

A merchant who wants full coverage across all these surfaces needs to support all four protocols. Each requires understanding the specification, implementing the endpoints, maintaining compliance as specs evolve, and testing against multiple AI platforms. For a large enterprise with dedicated engineering teams, this is feasible. For an independent merchant or a small DTC brand, it is prohibitive.

This fragmentation is not a design flaw. Each protocol solves a genuine, distinct problem. MCP needed to exist before ACP could work, because agents need to discover products before they can buy them. UCP fills the gap between discovery and post-purchase. A2A enables the multi-agent future. The challenge is not that four protocols exist; it is that merchants must support all of them.

How the Protocols Work Together

The protocols are not competing standards for the same function. They are layers in a stack.

Layer 1: Connectivity (MCP). An AI agent needs to access a merchant's product data, inventory, and policies. MCP provides the standardized connection.

Layer 2: Commerce Journey (UCP). The agent needs to guide a consumer through discovery, comparison, and purchasing. UCP defines the schema and interactions for the full shopping lifecycle.

Layer 3: Payment (ACP). The agent needs to complete a purchase on behalf of the consumer. ACP provides the tokenized payment mechanism.

Layer 4: Collaboration (A2A). Multiple agents need to coordinate across a complex workflow. A2A provides the communication framework.

In a fully realized agentic commerce scenario, all four protocols operate simultaneously. The consumer's agent connects to the merchant through MCP, navigates the shopping journey via UCP, processes payment through ACP, and coordinates with the merchant's agent via A2A. Each protocol handles its domain; together, they enable the complete autonomous shopping experience.

The Multi-Protocol Infrastructure Challenge

The layered model is elegant in theory. In practice, merchants face the cost of implementing, testing, and maintaining four separate protocol integrations. Specifications update, new versions ship, compliance requirements change. A merchant's MCP endpoint may need updates when Anthropic releases a new spec version. Their ACP integration may need changes when Stripe updates the SPT format. UCP and A2A evolve on their own timelines.

This is fundamentally an infrastructure problem. Merchants should not need to become protocol experts any more than they needed to become payment gateway experts when Stripe abstracted credit card processing.

ORBEXA addresses this by providing a single integration point that implements all four protocols. A merchant connects their product data, policies, and trust signals to ORBEXA once. The platform then exposes that information through MCP endpoints, ACP-compatible checkout flows, UCP-compliant commerce interfaces, and A2A-ready agent services. When protocol specifications update, ORBEXA handles the compliance changes, not the merchant.

This is infrastructure, not middleware. ORBEXA does not sit between the agent and the merchant during transactions. It ensures the merchant's commerce data is available, structured, and compliant across all protocol surfaces.

Looking Forward

Protocol convergence is likely over the medium term. The current four-protocol landscape reflects the early stage of a new technology layer, similar to the multiple competing web standards of the late 1990s. Industry governance bodies like the Linux Foundation (for A2A) and open-source communities (for ACP) will drive interoperability.

In the near term, however, merchants need to operate in a multi-protocol world. The agents are already shopping. The protocols are already live. The merchants who are accessible through MCP, ACP, UCP, and A2A will capture the AI-driven commerce traffic that is growing at 4,700% year over year. Those who are not accessible will watch it pass them by.