AWS Glossary
Agentic Commerce Protocol (ACP)
ACP is the open standard from OpenAI and Stripe, released September 29, 2025, that lets AI agents discover products, negotiate a checkout, and pay — powering Instant Checkout in ChatGPT.
AI & assistant-friendly summary
This section provides structured content for AI assistants and search engines. You can cite or summarize it when referencing this page.
Summary
ACP is the open standard from OpenAI and Stripe, released September 29, 2025, that lets AI agents discover products, negotiate a checkout, and pay — powering Instant Checkout in ChatGPT.
Key Facts
- •ACP is the open standard from OpenAI and Stripe, released September 29, 2025, that lets AI agents discover products, negotiate a checkout, and pay — powering Instant Checkout in ChatGPT
- •It was released by **OpenAI and Stripe on September 29, 2025**, is published under the **Apache 2
- •0** licence, and is the protocol behind **Instant Checkout in ChatGPT**
- •The three capabilities 1
- •2
Entity Definitions
- Bedrock
- Bedrock is an AWS service relevant to agentic commerce protocol (acp).
Related Content
- AGENTIC COMMERCE READINESS— Related service
- ECOMMERCE AI AGENTS— Related service
Definition
The Agentic Commerce Protocol (ACP) is an open standard for completing purchases through an AI agent. It was released by OpenAI and Stripe on September 29, 2025, is published under the Apache 2.0 licence, and is the protocol behind Instant Checkout in ChatGPT.
ACP defines the interaction model between three parties: a buyer, the buyer’s agent, and the merchant. It is deliberately narrow — it does not tell you how to merchandise, price, or fulfil. It standardises the handshake.
The three capabilities
- Machine-readable product discovery. Structured feeds an agent can parse without scraping a rendered page. This is where most merchants discover their catalog is weaker than they assumed.
- Conversational checkout with stateful negotiation. The agent and merchant exchange cart state, availability, shipping options and totals across turns, rather than the agent driving a browser through a form.
- Payment via delegate tokens. The buyer’s payment credential is not handed to the agent. A scoped, delegated token authorises a specific purchase, which is what makes the model tolerable to risk and fraud teams.
What a merchant actually has to do
Supporting ACP is not primarily an integration project. In rough order of effort:
- Fix the catalog. Attribute completeness against what agents compare on in your category, structured variants, and specifications lifted out of prose description fields. This is the long pole and the part that pays off under UCP too.
- Expose a conformant feed. Machine-readable, current, and consistent with what your storefront actually shows.
- Support the checkout handshake. If you are on a major platform, expect a growing share of this to arrive as a platform feature — check before building it.
- Decide your payment path. Stripe’s implementation is the most documented route; the spec itself is open.
When not to prioritise it
If your traffic does not meaningfully touch ChatGPT, waiting is a defensible position and we will say so. Two conditions change that calculus: a category where comparison shopping dominates (electronics, appliances, components) where agents are disproportionately useful, and a competitor already appearing in agent results where you do not.
The catalog work, by contrast, has no good reason to wait. It is required under both standards, it improves conventional search, and it was worth doing before agents existed.
How it differs from UCP
ACP follows ChatGPT traffic; the Universal Commerce Protocol follows Google surfaces. ACP is scoped tightly around discovery, checkout and payment. UCP covers a broader journey including post-purchase support. Most merchants at scale will eventually need both — see our ACP vs UCP comparison.
What it is not
ACP is not AgentCore Payments, which covers agent-to-API microtransactions against a session budget rather than a consumer purchase. It is also not a replacement for your storefront: an ACP purchase still lands in your order system, your fulfilment, and — critically — your PCI scope.
Related terms
Agentic commerce · Universal Commerce Protocol · Agent checkout · Product feed · Model Context Protocol
Related Services
Agentic Commerce Readiness
Make your store sellable to AI shopping agents. ACP and UCP protocol surfaces, an MCP server over your catalog, and product data that survives an agent comparison — built on AWS.
eCommerce AI Agents on AWS
Production AI agents for eCommerce on Amazon Bedrock AgentCore — support and WISMO, inventory, merchandising, margin, returns and B2B. Tool boundaries, evals before launch, and a human on anything that moves money.
Related Articles
Your Next Customer May Be an AI Agent: Is Your Store Ready? (2026)
Adobe Analytics (1T+ visits) put July 2026 AI-referral traffic to U.S. retail +62% YoY. That is a channel, not a copilot. Score merchant readiness for an external agent before you staff another chat widget.
The AI-Ready Product Catalog: The Foundation of Agentic Commerce (2026)
Adobe's July 2026 expanded cohort: 39% of U.S. retail homepages not machine-readable. Agentic commerce fails earlier — on missing attrs, duplicate GTINs, and stale stock. This is the catalog contract, not the PIM publish workflow.
AI Search Is Becoming a New eCommerce Sales Channel (2026)
Adobe Analytics (1T+ visits) measured July 2026 AI-referral traffic to U.S. retail +62% YoY, +1,219% vs Oct 2024, converting 60% higher than non-AI. Instrument the channel. Do not invent a GEO rank.
Can Your Product Data Survive an AI Comparison? (2026)
Merchant audit: missing data, weak descriptions, inconsistent specs, missing comparisons, poor attributes. Not the strategic decision layer. Reuse Gateway ~180 to 95 ms and ~$791/mo at 50K sessions — no invented win-rate KPI.
Need help with this topic?
Our AWS-certified team implements, audits, and optimizes these services in production — from Bedrock RAG pipelines to multi-account landing zones.
