From A2A to UCP: All the Relevant Protocols for Your Agentic Commerce Operations

If you’re building commerce infrastructure for AI agents, you’re navigating a landscape of emerging protocols that sound like alphabet soup: MCP, A2A, ACP, UCP, onX, SCP, UIP and so on. Each serves a specific function in the agentic commerce stack, but understanding how they layer together determines whether your infrastructure supports AI-driven shopping or fights it.

This guide explains most major protocols currently shaping agentic commerce, how they fit together, and what each one enables. Whether you’re evaluating which standards to implement or trying to make sense of announcements from Shopify, Google, OpenAI or Anthropic, this is your reference.

New to Model Context Protocol? MCP is APIs for the AI era. This guide assumes basic familiarity with MCP as the foundation. If you need the fundamentals first, read our complete MCP explainer before continuing. 

The Protocol Stack: How Standards Layer in Agentic Commerce

Before diving into individual protocols, understanding the stack architecture matters. These protocols don’t compete, they layer:

Foundation Layer: A2A, MCP Model Context Protocol exposes data and functionality from any system to AI models. Think of it as the API layer for AI.

Selling Channel Layer: ACP, UCP Application-to-application protocols that define how AI agents interact with commerce systems during product discovery, checkout, and payment capture.This also includes specialized protocols that extend core functionality for specific use cases like shopper identity or loyalty programs.

Payment Channel Layer: AP2 Application-to-payment protocols that define how AI agents and commerce systems interact with payment providers during transaction processing, authorization, and settlement. This layer standardizes payment instrument selection, fraud detection workflows, and transaction lifecycle management across traditional and emerging payment methods.

Order Management / Fulfillment Layer: onX Order Network eXchange exposes inventory, fulfillment, and order capture capabilities without dictating workflows.

To better understand the relationship between onX and UCP read our guide here. Now let’s examine each protocol.

Foundation: A2A and MCP

Agent-to-Application Protocol (A2A)

What it is: An early standard for enabling AI agents to interact with commerce applications programmatically.

What it does: Defines callbacks and workflows for AI agents to browse catalogs, check inventory, and complete transactions with minimal human intervention.

What it doesn’t do: Provide the level of workflow specificity that newer protocols like UCP deliver. A2A established concepts but left implementation details open.

Used for: Foundational agent-commerce interactions in early agentic shopping implementations.

Current status: Largely superseded by more opinionated protocols (ACP, UCP) that learned from A2A’s patterns.

Model Context Protocol (MCP)

What it is: An open standard (proposed by Anthropic in November 2024) that allows AI models to access data and functionality from external systems through a standardized interface.

What it does: Provides the fundamental mechanism for AI agents to query databases, trigger actions in business SaaS, and retrieve information that wasn’t part of their training data.

What it doesn’t do: Dictate commerce workflows or define how shopping experiences should work. MCP is infrastructure, not application logic.

Used for: Exposing any system’s data to AI “heads”. In commerce specifically, this means making inventory levels, order data, customer information, and fulfillment capabilities accessible to AI agents.

Technical reality: MCP defines how to expose resources (data), tools (actions), and prompts (instructions) through a standardized server interface. AI models connect as clients and interact with these resources using the protocol.

For a deep dive into MCP and its implications for commerce operations, read our complete guide: What is MCP? The Missing Link Between AI and Your Business Data

Selling Channel Standards: ACP, and UCP

Two protocols compete (or coexist, depending on adoption) to standardize how AI agents handle product discovery and checkout. Understanding the differences matters.

Agent Commerce Protocol (ACP)

What it is: Google’s standard for AI-powered shopping, announced as part of their broader AI commerce strategy.

What it does: Defines structured workflows for how AI agents should handle product search, inventory queries, checkout flows, and payment capture when shopping through Google’s AI surfaces. It enables payment to be taken outside of the commerce platform by the payment provider and then have the order flow to the commerce platform pre-paid.

What it doesn’t do: Dictate back-office operations like fulfillment or returns. ACP focuses on the selling channel side, relying on callbacks to other systems for operational data.

Used for: Enabling products to be discoverable and purchasable through Google AI agents (Gemini and future commerce surfaces).

Technical details: ACP provides specific callback points where merchants must surface data (inventory, pricing, promotions) and accept actions (order capture). Implementations must support these callbacks to participate in Google’s agentic commerce ecosystem.

Universal Commerce Protocol (UCP)

What it is: Shopify’s open standard for AI-driven commerce, designed to work across any AI platform (not just Shopify’s).

What it does: Defines comprehensive workflows for product discovery, checkout, and payment capture. UCP is highly opinionated about how these flows should work, providing detailed specifications for each step.

What it doesn’t do: Originate data like inventory levels or fulfillment capabilities. UCP defines callback points where AI agents retrieve this information from external systems.

Used for: Enabling commerce across multiple AI platforms while maintaining consistent workflows. UCP aims to be the standard that works with Claude, ChatGPT, Gemini, and future AI agents.

Technical details: UCP specifies exact callback sequences for shopping flows. For example, when an AI agent needs inventory data, UCP defines how that callback should be structured and what data format to expect in return.

Relationship to onX: UCP callbacks often reach back to onX-compliant systems for operational data. When a UCP-powered AI agent asks “is this in stock?”, it’s calling an onX inventory endpoint for the answer.

Payment Channel Layer

AP2 (Agent Payment Protocol 2)

What It Is: AP2 is an open protocol that standardizes how AI agents interact with payment systems during commerce transactions. It defines the data structures, authentication methods, and workflow patterns that allow AI agents to securely initiate payments, handle authorization requests, manage payment instrument selection, and process transaction outcomes on behalf of users.

What It Does: AP2 enables AI agents to execute payment workflows that previously required direct human interaction with payment interfaces. The protocol supports multi-step payment flows including instrument tokenization, risk assessment coordination, 3D Secure authentication, split payment orchestration, and failed transaction recovery. AP2 allows agents to present payment options based on merchant acceptance, user preferences, and transaction context while maintaining PCI compliance through standardized token exchange rather than handling raw payment credentials. The protocol also defines how agents should handle edge cases like partial authorizations, currency conversion, and multi-party payment splitting.

What It Doesn’t Do: AP2 does not store payment credentials, process card transactions, or replace payment gateways and processors. It does not handle the cryptographic operations required for payment security or manage the actual movement of funds between accounts. AP2 does not define fraud detection algorithms, determine merchant acceptance rules, or establish interchange rates. The protocol does not bypass existing payment network requirements like PCI-DSS compliance or strong customer authentication mandates. AP2 is a coordination layer that allows agents to interact with existing payment infrastructure, not a replacement for that infrastructure.

Operations Layer: Order Network eXchange (onX)

What it is: An open standard (co-founded by Pipe17 through the Commerce Operations Foundation) that exposes inventory, order capture, and fulfillment capabilities through MCP.

What it does: Provides standardized access to back-office commerce operations without dictating workflows. onX defines what data should be available (inventory levels, order status, fulfillment capacity) but not how that data gets used.

What it doesn’t do: Define shopping experiences or checkout flows. onX handles everything after the “buy” decision, while protocols like UCP handle everything before it.

Used for: Connecting AI agents to order management systems, warehouse management systems, and fulfillment networks. When a UCP or ACP agent needs operational data, it calls onX endpoints.

Technical details: onX exposes resources through MCP servers. An AI agent (or a UCP callback) can query inventory across multiple fulfillment locations, capture orders, check fulfillment capacity, or initiate returns by calling standardized onX endpoints.

Why this matters: onX provides the operational foundation that makes selling-channel protocols (UCP, ACP) work. Without standardized access to inventory and fulfillment data, AI agents can’t make accurate promises during checkout.

Learn more

Extension Protocols: Specialized Standards

Beyond the core stack, extension protocols handle specific commerce functions that require their own standards.

Shopper Context Protocol (SCP)

What it is: An MCP-based protocol for portable shopper identity and preferences across commerce experiences.

What it does: Functions like a universal passport that shoppers present to new commerce experiences. Contains profile data, preferences, purchase history, and permissions.

What it doesn’t do: Replace traditional authentication. SCP provides context enrichment, not identity verification.

Used for: Enabling AI agents to provide personalized shopping experiences across different merchants without requiring shoppers to re-enter preferences each time.

Technical analogy: Similar to OAuth token grants, but for shopper context rather than API access. When you arrive at a new merchant’s site, you grant access to your SCP data, and the AI agent can immediately personalize recommendations.

Unified Incentives Protocol (UIP)

What it is: An extension of UCP specifically designed to handle loyalty programs, rewards, and promotional incentives in AI-driven shopping.

What it does: Standardizes how AI agents access, apply, and communicate loyalty rewards, points balances, and promotional offers during shopping.

What it doesn’t do: Replace merchant-specific loyalty programs. UIP provides a standard interface for AI agents to interact with whatever loyalty system a merchant runs.

Used for: Enabling AI agents to answer questions like “can I use my points for this purchase?” or “what rewards am I earning?” without merchants building custom integrations for every AI platform.

Future extensions: Expect more UCP and ACP extensions as specific use cases emerge. Subscription management, gift card handling, and trade-in programs are all candidates for standardized extension protocols.

How These Protocols Work Together: A Complete Transaction

Understanding protocols individually helps. Seeing how they interact during an actual transaction makes the architecture clear.

Step 1: Product Discovery (UCP/ACP Territory)

A customer asks an AI agent: “Find me running shoes under $150.”

The AI agent uses UCP workflows to query merchant catalogs. UCP defines how this query should be structured and what data format merchants should return.

Step 2: Inventory Check (UCP Callback to onX)

The customer asks: “Do you have size 10 in blue?”

UCP defines a callback point for inventory queries. That callback reaches an onX-compliant MCP server, which exposes real-time inventory across fulfillment locations. The AI agent receives inventory data through this standardized interface.

Step 3: Personalization (SCP)

The AI agent accesses the shopper’s context through SCP: previous purchases, size preferences, shipping address, preferred payment method.

This enables personalized recommendations without the shopper re-entering information.

Step 4: Checkout (UCP/ACP)

The customer decides to buy. UCP defines the checkout workflow: collecting shipping information, calculating tax, applying promotions.

Step 5: Loyalty Application (UIP)

During checkout, the AI agent queries UIP to check available rewards: “You have 500 points worth $5. Apply to this purchase?”

Step 6: Order Capture (UCP to onX Handoff)

Payment is captured. UCP’s job ends. The completed order flows to an onX endpoint for fulfillment processing.

From this point forward, onX showcases all operational work: order routing, fulfillment, shipment tracking, returns.

Step 7: Ongoing Operations (onX via MCP)

AI agents can continue querying order status, initiating returns, or checking delivery estimates by calling onX endpoints through MCP.

Implementation Priorities: Which Protocols Matter When

Not every protocol requires immediate implementation. Prioritization depends on your commerce strategy.

Essential for All Commerce:

• MCP: Foundation layer. Without this, nothing else works.
• onX: Operations layer. Required for AI agents to access inventory and capture orders.

Essential for AI-Driven Selling:

• UCP or ACP: Choose based on which AI platforms you’re targeting. UCP for cross-platform, ACP for Google-specific.

Value-Add Extensions:

• SCP: Implement when personalization across AI agents becomes a competitive differentiator.
• UIP: Implement when loyalty programs are strategic and you want AI agents to leverage them.

Experimental/Early:

• A2A: Consider this historical context rather than an active implementation target.

Why This Matters for Commerce Infrastructure

These protocols aren’t academic exercises. They determine which commerce systems can participate in AI-driven shopping.

If your inventory system can’t expose data through onX, AI agents can’t accurately promise availability. If your commerce platform doesn’t support UCP or ACP, your products won’t be discoverable through AI shopping experiences. If your loyalty program can’t integrate with UIP, AI agents can’t help customers leverage rewards.

The infrastructure you build (or buy) today determines whether you capture AI commerce opportunities or watch them flow to competitors with better protocol support.

At Pipe17, we’re implementing these standards because we see how they work together to enable the next phase of commerce. MCP exposes our order operations capabilities. onX standardizes how AI agents access them. UCP/ACP integration means orders flow seamlessly from AI shopping experiences into fulfillment.

That’s the complete stack. That’s what enables agentic commerce at scale.

---

Want to see how Pipe17 implements these protocols to enable AI-powered commerce? Book a demo to explore our MCP server, onX implementation, and protocol-ready order operations platform.