Architecture

Agentic Commerce Protocol (ACP): What Merchants Need to Know

Learn what the Agentic Commerce Protocol is, where ACP fits in AI checkout, and how merchants should compare it with broader commerce protocols.

Updated May 17, 2026

The Agentic Commerce Protocol is the protocol layer Stripe documents for enabling in-context selling on AI agents. In practical terms, ACP is about checkout inside an agent experience: a user discovers a product in an AI surface, keeps the purchase flow in context, and the merchant connects to that flow without rebuilding the whole storefront around a human browser session.

What official docs say#

Stripe’s agentic commerce documentation tells businesses to enable in-context selling on AI agents using the Agentic Commerce Protocol. Stripe’s launch announcement also describes ACP as an open standard codeveloped with OpenAI. That places ACP squarely in the transaction experience layer rather than the full search, product-data, or multi-agent orchestration stack.

Primary sources:

ACP compared with nearby protocols#

ProtocolMain job
ACPIn-context checkout for AI-agent experiences
UCPBroader commerce lifecycle and capability negotiation
MCPTool and data access for agents
x402Machine payment flow for HTTP resources

The UCP vs ACP vs MCP guide covers the broader comparison. ACP is best understood as a focused commerce-experience protocol, not a replacement for every other part of an agent-ready stack.

When ACP matters#

ACP becomes relevant when:

  • buyers discover products inside AI conversations
  • merchants want checkout to happen without a clumsy redirect path
  • the AI surface is becoming a real sales channel
  • the merchant already has product data, pricing, and policy information clean enough for automated buying

That last point is easy to underestimate. Checkout does not rescue bad product data. The ChatGPT product recommendations SEO checklist explains what needs to happen before selection.

What merchants should prepare first#

Readiness areaWhy it matters
Product dataAgents need facts before they recommend or buy
Price and availabilityStale data breaks trust immediately
Returns and shippingBuyer constraints often include both
Fraud controlsAgent-driven orders still need risk handling
Order confirmationThe result must be deterministic and auditable

ACP may reduce friction at checkout, but the execution layer still needs reliable backend state transitions.

ACP vs traditional ecommerce checkout#

Traditional checkoutACP-style flow
Buyer navigates storefrontBuyer may stay in the AI interface
Session and UI dominateProtocol and structured state dominate
Conversion depends on page flowConversion depends on context plus data quality
Human reads all detailsAgent may pre-filter options

This is why old checkout analytics alone are not enough. Merchants need to track discovery, recommendation, handoff, checkout completion, and post-purchase correctness as one chain.

Risks and limits#

Do not treat any protocol as automatic revenue. ACP depends on:

  • platform adoption
  • merchant integration quality
  • user trust
  • correct product information
  • checkout policies that still make sense when an agent acts for a buyer

For regulated products or high-consideration purchases, human confirmation may remain necessary even if the checkout surface becomes agent-native.

FAQ#

Is ACP the same thing as agentic commerce?#

No. Agentic commerce is the broader market shift. ACP is one protocol approach for in-context selling on AI agents.

Does ACP replace UCP?#

No. ACP is focused on checkout experience, while UCP covers a wider commerce lifecycle.

Should every merchant implement ACP now?#

Not automatically. First confirm that AI-assisted shopping is a meaningful channel for your category and that your data foundation is ready.

What is the best first step?#

Fix product data, product schema, policy clarity, and checkout determinism before adding new protocol surfaces.

Bottom line#

ACP matters because checkout is moving closer to the conversation. Merchants that want to sell through AI surfaces need more than good copy: they need clean product facts, reliable operations, and checkout flows built for machine mediation.