The Agentic Commerce Protocol Stack: MCP, WebMCP, ACP, UCP, AP2, and A2A

Agentic commerce is a macro ecosystem, not one protocol. Different standards address how agents discover merchant information, call capabilities, build commerce interactions, authorize payment, and communicate with other agents.

The stack is useful because the acronyms are often compared as if they were substitutes. They are not. A merchant may encounter several layers through platforms and providers without implementing any protocol directly.

The five-layer map

Layer 5 — Agent-to-agent communication
  A2A

Layer 4 — Payment authorization
  AP2

Layer 3 — Commerce and checkout
  ACP, UCP (with some discovery capabilities spanning layers)

Layer 2 — Tool execution
  MCP, WebMCP

Layer 1 — Discovery
  Structured product data, schemas, feeds, catalogs, platform discovery surfaces

The model is explanatory rather than a formal standards taxonomy. Its job is to keep unlike functions separate.

Layer 1: discovery

Before an agent can consider a merchant, it needs accurate, accessible information about the merchant and its products. Discovery can involve structured data, platform catalogs, merchant feeds, ordinary crawlable pages, and protocol capabilities that carry product information. OpenAI expanded ACP to support product discovery in March 2026, while Shopify Catalog supplies product data to supported AI channels for eligible merchants. A protocol can therefore span more than one explanatory layer without making the layers interchangeable.

Discovery does not imply action or transaction. A product appearing in an AI answer is different from an agent being authorized to modify a cart or complete a purchase.

Questions for operators:

Which system is the source of truth for product information?
Which channels receive the data?
How are availability, price, policy, and product changes synchronized?
Which claims describe crawler access versus platform syndication?

Use The AI Shopping Product Data Checklist for the merchant-side data contract rather than treating a protocol name as evidence of complete product information.

Layer 2: tool execution

MCP

Anthropic introduced the Model Context Protocol in November 2024. MCP lets agents connect to declared tools and data sources through MCP servers.

In December 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation. Anthropic said the project’s maintainer-led governance model remained unchanged.

WebMCP

WebMCP is an experimental browser-side proposal from the W3C Web Machine Learning Community Group. Its current imperative API uses document.modelContext to let a connected document expose structured tools.

WebMCP is inspired by MCP vocabulary but is a distinct web API, not MCP running in the browser. Its API and implementation status must be checked immediately before publication or deployment.

MCP and WebMCP both concern callable capabilities, but they expose them in different environments. See WebMCP vs. MCP for the decision framework.

Layer 3: commerce and checkout

ACP

OpenAI and Stripe are ACP’s founding maintainers. ACP began with commerce concepts including checkout sessions, carts, feeds, delegated payment, and delegated authentication. OpenAI’s March 2026 product-discovery release expanded ACP’s role: merchants can provide product feeds and promotions through ACP delivery paths so ChatGPT can use more complete product information during discovery.

Checkout behavior is not uniform. OpenAI says it is emphasizing product discovery and merchant flexibility to use their own checkout, while its current shopping documentation says some eligible products and merchants may still show Instant Checkout. Shopify currently describes ChatGPT as a discovery-focused referrer for eligible Shopify stores, with purchase completed through the merchant checkout in an in-app browser or new tab. Verify the merchant, channel, and date before describing a checkout path.

For current merchant participation paths, see How to Get Your Products Into ChatGPT Shopping.

UCP

At its January 2026 launch, Google said UCP was built with Shopify, Etsy, Wayfair, Target, and Walmart and endorsed by more than 20 other organizations. The contributor ecosystem has expanded, so distinguish the launch group from current governance.

UCP describes a broader commerce interaction model covering stages beyond checkout. It is designed to compose with other agent and payment infrastructure, but each production implementation needs separate verification.

ACP and UCP overlap in commerce, but their scopes and ecosystem integrations differ. “Which protocol wins?” is less useful than “Which channel and merchant capability are we trying to support?”

Layer 4: payment authorization

AP2 addresses authorization and auditability for agent-led payments. It uses signed mandates to represent what a user authorized and under which constraints.

AP2 is not the same thing as a product feed, shopping conversation, or checkout session. Do not infer supported payment rails or production readiness without checking current official documentation.

Layer 5: agent-to-agent communication

A2A defines patterns for agents to discover and communicate with other agents. Google originated the project and donated it to the Linux Foundation in June 2025.

This layer matters when separate agents coordinate work. It should not be used as a label for a shopper-facing chat interface.

How an ecommerce operator should use the map

When evaluating a platform or vendor, ask:

Which layer and primary job does the product actually address?
Where does the interaction occur: an external channel, a chat-contained flow, a rebuilt storefront, or the merchant’s existing interface?
What state or product data is available, and how current is it?
Which actions are supported, by whom, and under what permissions?
How is an action or transaction verified?
Which protocol claims are live implementations versus roadmap compatibility?

The protocol name is not a substitute for an implementation artifact.

Where kn8 fits

kn8 is an AI shopping agent whose workspace is a connected storefront interface. The host supplies current state and visible targets; kn8 guides through that interface, invokes only supported actions, and verifies what changed. That product model is separate from protocol compatibility, which should be claimed only from a current implementation artifact. See the bounded product model →

Key takeaways

Agentic commerce spans several functional layers.
Discovery, tool execution, commerce, payment authorization, and agent communication are different jobs.
ACP and UCP should not be reduced to a timeless winner-versus-loser comparison, and their current discovery and transaction scopes must be checked before reuse.
WebMCP is an experimental browser proposal distinct from MCP.
Protocol compatibility must be demonstrated, not inferred from conceptual similarity.