Stripe ACP: The Agentic Commerce Protocol
An analysis of Stripe Agentic Commerce Protocol (ACP), enabling autonomous AI agents to interact with e-commerce systems.
Introduction to Stripe ACP
The Agentic Commerce Protocol (ACP) by Stripe is a standardized API and payload protocol designed to bridge the gap between traditional commercial web platforms and autonomous purchasing agents. In the modern web ecosystem, checkout flows are heavily optimized for human users, relying on visual cues, multi-step forms, and interactive anti-fraud challenges. These designs present significant obstacles to AI agents attempting to purchase physical goods or register for services on behalf of their users.
Stripe ACP solves this issue by exposing a machine-readable checkout interface alongside standard web pages. Merchants integrating the protocol declare their products, pricing structure, inventory status, and payment rules in a structured format that agents can consume programmatically. By standardizing the communication channel, Stripe ACP enables seamless automated purchasing while maintaining strict security controls and fraud prevention measures.
Protocol Mechanics and Session Initiation
An agentic commerce session begins when the purchasing agent requests the ACP metadata endpoint for a specific merchant or product. The merchant returns an ACP manifest containing detailed schema definitions for the checkout requirements. These details include supported payment methods, shipping options, required billing fields, and any conditional rules, such as geographic restrictions.
The agent parses the manifest and compiles the necessary user information from its secure local vault. To finalize the purchase, the agent submits a structured payload containing the selected items, shipping address, and a delegated payment credential. Stripe ACP processes the request, performs risk evaluation, and returns a transaction confirmation message containing tracking information and receipt details.
The table below outlines the core properties of an ACP manifest payload:
| Property | Type | Description |
|---|---|---|
merchantId | String | The unique identifier of the merchant within the Stripe network. |
supportedCurrencies | Array | A list of currency codes accepted for the checkout session. |
requiredFields | Array | The specific client details (e.g., shipping, phone) required for compliance. |
taxExemptStatus | Boolean | Indicates whether the merchant supports tax exemption for automated buyers. |
Delegated Checkout Implementation
Integrating Stripe ACP requires server-side route handling to negotiate and settle payment intents. Below is a simplified representation of how an agent initiates a delegated checkout session using Stripe ACP endpoints:
interface AcpCheckoutPayload {
merchantId: string;
items: Array<{ id: string; quantity: number }>;
shippingAddress: object;
paymentMethodId: string;
}
async function initiateAcpCheckout(payload: AcpCheckoutPayload): Promise<object> {
const response = await fetch('https://api.stripe.com/v1/acp/checkouts', {
method: 'POST',
headers: {
'Authorization': `Bearer ${process.env.STRIPE_SECRET_KEY}`,
'Content-Type': 'application/json',
'Stripe-Version': '2025-05-15'
},
body: JSON.stringify(payload)
});
if (!response.ok) {
const errorDetails = await response.json();
throw new Error(`Stripe ACP Checkout Failed: ${JSON.stringify(errorDetails)}`);
}
return response.json();
}By providing a clean programmatic path for commerce, Stripe ACP eliminates checkout friction and enables the next generation of transactional AI.
Latest Articles
View allDeploying Model Context Protocol (MCP) Servers for Agent Discovery
Deploy Model Context Protocol (MCP) servers over Streamable HTTP and stdio for secure tool exposure and agent discovery.
Enforcing Walled Garden Guardrails: SAP API Policies and Adobe CX
Enforce enterprise guardrails across SAP API policies and Adobe CX to prevent data exfiltration, token leakage, and unauthorized agent actions.
Agent-to-Agent (A2A) Interoperability and UI Handoffs
Master A2A protocol specifications and AG-UI handshakes for cross-domain agent task delegation, UI state coordination, and human oversight.