Agentic Commerce Protocol (ACP)

The Agentic Commerce Protocol (ACP) is an open standard that defines how an AI agent can complete a purchase from a business on a buyer’s behalf. It was released by OpenAI and Stripe in September 2025 under the Apache 2.0 licence, and Stripe’s documentation now lists Meta as a co-creator. MLforSEO’s Agentic SEO course describes it as the fast lane for being purchased inside an AI chat.

How it works

Stripe’s documentation describes ACP as a set of composable building blocks:

• Agentic checkout: creating, updating and completing checkout sessions, including cart, fulfilment options and payment.
• Cart and feed: letting agents browse product catalogues and manage carts.
• Delegated payment: passing scoped payment tokens between buyer, agent and business, so raw card details are not exposed to the agent.
• Delegated authentication: OAuth 2.0 so an agent can act for a buyer with a business.
• Orders and webhooks: updates for confirmation, shipping, delivery and refunds.

The business stays the merchant of record and keeps control of its checkout logic, pricing and payment provider. The specification is versioned on GitHub, with OpenAPI and JSON Schema definitions (the current Stripe docs reference a 2026-04-17 version).

How it’s used in SEO and marketing

In MLforSEO’s three agent types (research, recommendation, action), ACP serves the action agent, the one that buys, books or enrols. It sits in an agent protocol stack alongside WebMCP for in-browser actions and the Universal Commerce Protocol (UCP) for complex retail flows. For most businesses the immediate priority is the shared precondition: accurate, machine-readable product data (prices, availability, policies) in structured data and feeds, because an agent that cannot verify a required attribute moves on to a competitor.

Example

At launch in September 2025, ChatGPT’s Instant Checkout used ACP to let US users buy from Etsy sellers inside the chat. In March 2026 OpenAI said Instant Checkout was “moving to Apps” and that it was prioritising search and product discovery; ACP stayed as the infrastructure connecting ChatGPT users to merchants, now mainly through retailer apps where merchants control checkout.

Limitations and things to watch

• It is young and moving. Product names, partners and flows have already changed once; check the official specification before planning work.
• Implementation is a developer and payments project, usually through Stripe or another compatible provider, and availability depends on the AI platform and region.
• Adoption is sector-dependent, as the course stresses: you do not implement ACP in a day, and clean product data pays off whether or not you adopt it.

Learn it on MLforSEO

• Your audience is no longer only human introduces action agents and the protocols they look for.
• Agent Readiness Stack Audit is a free template for the entity, attribute and action layers.
• Consistency beats completeness: schema for machines, not results covers the product data agents rely on.
• The AI Search Optimisation & Agentic SEO course covers the full agent protocol stack in lesson 3.4.