Particular Audience Particular Audience
Talk to us

The retail operating system

One engine for every retail decision.

Most retailers run four or five separate tools.

Search in one tool, pricing in another, ads bolted on top — nobody owns the order shoppers see, and margin leaks between them. The Unified Decision Engine governs relevance, inventory, margin and sponsored demand in one pass, on every surface, including the AI assistants now doing the shopping.

Start with one surface. Leave the rest of your stack untouched.

Need one thing first? Start with our AI product recommendation engine or AI site search.

ACP, UCP and MCP: Retail Commerce Protocols Compared

Compare ACP, UCP and MCP by the job each performs. Separate commerce capabilities, tool connections, platform access and retailer implementation requirements.

ACP, UCP and MCP: what retailers should compare

ACP, UCP and MCP address related but different parts of an assistant-enabled commerce system. MCP connects applications to tools and context. ACP and UCP describe commerce interactions and capabilities. Compare the supported version, merchant responsibilities and platform integration before choosing an approach; the acronyms are not a list of interchangeable shopping channels.

Compare the job, not just the acronym

MCP, the Model Context Protocol, provides a general connection between AI applications and external tools or information. It does not define which products a shopping service must rank or whether a merchant has permission to transact through that service.

ACP, the Agentic Commerce Protocol, addresses commerce flows. Its supported integration can involve APIs or MCP. OpenAI separately documents its commerce onboarding, so a protocol implementation should not be confused with approval for a particular platform integration.

UCP, the Universal Commerce Protocol, defines commerce capabilities and supports multiple transports, including MCP. A useful architecture can therefore involve both a commerce protocol and MCP rather than requiring an exclusive choice between them.

Use a requirements matrix

Question — What to record — Why it matters

Which shopper job is supported? — Discovery, offer comparison, basket or transaction capability — Prevents a read interface being presented as checkout

Which contract applies? — Protocol version, supported capability and schema — Makes compatibility testable

Who accepts the transaction? — Merchant system, platform connection and commercial owner — Separates technical exchange from business responsibility

What is authorised? — Account, scope, user confirmation and expiry — Defines which actions may proceed

What happens on failure? — Error, retry, expiry and customer explanation — Prevents ambiguous outcomes

Fill the matrix with evidence from the intended service, not only the protocol’s generic specification. Two implementations using the same name can support different capability subsets or rollout stages. Record unsupported requirements plainly and decide whether a narrower journey is still useful.

Test the boundary between systems

Use a synthetic merchant scenario with a product lookup followed by an offer request. The lookup can succeed while the offer is unavailable in the shopper’s region. The correct result is a clear unsupported or unavailable response, not an invented delivery promise. If a later step can change a basket, test duplicate requests and confirmation independently from the successful lookup.

Keep protocol conformance, application behaviour and platform participation as separate acceptance records. A successful tool call proves that particular exchange worked; it does not prove the assistant will choose the tool, prefer the merchant or complete every shopping journey.

Place PA in the documented architecture

Use PA’s connector status to establish the current service boundary. A public connector demonstration can help explain terminology, while a retailer integration requires its own documented scope and tests. Agree which system owns product truth, ranking, offer freshness and any permitted action before presenting the architecture as available.

Illustrative example

SYNTHETIC: product lookup succeeds but the requested offer region is unsupported. The result records the unavailable offer without implying a purchasable or reserved item.

Sources

How Particular Audience does it

Frequently asked questions

Is UCP a replacement for MCP?

Not necessarily. UCP describes commerce capabilities and can use MCP as a transport. Compare the intended capability and binding rather than assuming the two names represent competing implementations of the same layer.

Does an ACP implementation automatically connect a retailer to ChatGPT shopping?

Do not assume so. Check the current platform onboarding, supported integration and merchant requirements separately. Generic protocol compatibility is not proof that a specific merchant or product is approved, eligible or displayed.

What evidence should an implementation partner provide?

Ask for the supported contract version, capability list, authentication boundary, sample exchanges and failure tests for your intended journey. Request separate evidence for platform participation and production deployment. A diagram or public sample tool is not a substitute for those records.

Related guides

  • What is agentic commerce?: Understand agentic commerce across discovery, comparison and authorised action. Map product data, permissions, checkout boundaries and useful retail measures.

Buyer questions · All guides