How Merchants Can Accept AI Agent Payments: An Agentic Commerce Checkout Guide

TL;DR

• You can accept AI-agent payments through your existing catalog, checkout, and payment provider. Add checks for agent identity and delegated authority before an agent places an order.
• Discovery comes first. Give agents current product, price, and availability data, then verify which agent represents the buyer and whether the buyer authorized that purchase. Skyfire’s KYA model links agent identity with a user mandate.
• Checkout orchestration should send the cart through a flow your platform supports. Before payment authorization, check the mandate’s merchant, product, and amount against the order, and apply your spend limits.
• Fraud and audit controls should send declined or out-of-scope orders for human review. Keep the agent identity reference, mandate, authorization decision, and payment reference so you can investigate disputes.

Is your stack ready for agent checkout

We recommend answering yes or no to each check before you start integration work. A “no” identifies a dependency to resolve, not necessarily a reason to replace your commerce stack.

• Catalog: Can an agent read structured product details, current prices, and availability? BigCommerce recommends structured data so agents can evaluate what you sell.
• Checkout APIs: Can your stack return live inventory and pricing, create an order, and report its status without relying on a shopper’s browser session?
• Payment gateway: Can your provider authorize the agent-initiated payment flow you plan to accept? Confirm the payment method and authorization requirements with that provider.
• Agent-traffic policy: Have you set rules for identifying agent requests, checking delegated purchase authority, and sending exceptions to a human?
• Vendor availability: Can your chosen vendor enable that flow for your store and location today? Stripe labels its Agentic Commerce Suite US-only, while Shopify describes select-merchant pilots.

Five yes answers establish readiness for the specific flow you checked. They do not establish support for every agent protocol.

Catalog and discovery: making products legible to agents

An agent needs product data it can read and a way to confirm that an item remains purchasable. Publish structured records with product and variant identifiers, prices, currency, availability, and fulfillment details. We recommend letting the agent request a current price and inventory check before checkout. A product listing may be accurate when an agent discovers it but stale by the time the agent places an order.

Merchants can expose that data through different platform mechanisms. Shopify describes UCP as covering product discovery and checkout, while its Agentic Storefronts connect catalog data to specific AI sales channels. Stripe lets eligible merchants upload a catalog in its Dashboard and choose agent channels for its US-only Agentic Commerce Suite. BigCommerce points merchants to its API-first storefront and catalog APIs, along with structured product data and real-time availability.

Catalog access lets an agent find an item and check its terms. It does not establish which agent is requesting the purchase, whom the agent represents, or how much that person authorized it to spend.

Verifying agent identity and delegated authority

A merchant needs to establish which agent submitted an order, which person authorized it, and whether that authorization covers the proposed purchase. A valid login alone does not answer whether an agent may buy a particular product at a particular price. At checkout, you should compare the agent’s identity and delegated authority with the merchant, item, and amount in the order. If the authority is missing, expired, or outside that scope, pause the order for human approval.

At Skyfire, our KYA token is one approach to identifying the actor. Skyfire describes the token as binding the platform, agent, and human principal to a user mandate. In Skyfire’s illustrated checkout flow, the agent requests a token for a merchant URL, and the mandate specifies the merchant, product, and amount before Agentic Wallet authorizes payment. Merchants should confirm how their own checkout will receive and validate those credentials. Skyfire’s example does not establish a universal token standard or guarantee acceptance at every merchant.

OAuth scopes can grant access to an application, and API keys can authenticate a caller. Enterprise IAM and non-human identity governance, including Okta-style tools, can manage credentials and permissions for bounded internal workflows. For an autonomous agent purchasing across merchant sites, you also need a way to connect each proposed transaction to the agent and the user’s current authority. Those controls complement existing IAM rather than replace it. Agent identity records can support attribution and audit evidence, but they do not independently satisfy KYC requirements.

Orchestrating checkout across agent protocols

Merchants need a separate checkout path for each agent protocol they choose to accept. Before building one, map how the channel sends a cart, requests shopper approval, and passes payment authority to your existing checkout. A working connection to one channel does not establish support for another.

ChatGPT Instant Checkout uses the Agentic Commerce Protocol (ACP), with Stripe powering its payment flow. Stripe says merchants using its Agentic Commerce Suite remain the merchant of record and manage pricing and fulfillment. Shopify’s Universal Commerce Protocol (UCP) takes a different approach. Merchants and agents publish their supported capabilities, negotiate payment handlers, and can escalate a checkout for human approval. An ACP connection therefore cannot be assumed to handle a UCP checkout.

AP2 addresses proof of delegated payment authority through signed mandates. A protocol overview describes AP2 mandates and distinguishes x402, which serves machine-to-machine purchases such as paid API calls. We would keep x402 outside a storefront checkout plan unless the merchant also sells those resources.

For each supported path, the merchant should check the cart against current price and inventory, verify the applicable approval or mandate, and send an authorized payment request to its processor. If a channel cannot provide the required authority, the checkout should pause for human approval or decline. Skyfire describes an agent identity and user-mandate approach, but its cited first-party documentation does not confirm integrations with ACP, UCP, AP2, or x402. Merchants should evaluate that identity and mandate layer separately from protocol connectivity.

Authorizing the transaction and enforcing spend controls

A verified agent should receive payment authorization only when its proposed order fits the buyer’s delegated authority. Before submitting a charge, your checkout should compare the agent identity and mandate with the merchant, product, and final amount, including tax and shipping. A valid payment credential alone cannot establish which agent or task initiated the purchase.

Payment providers can carry the charge through different paths. Stripe describes Shared Payment Tokens for agent-initiated payments with associated risk signals and separately offers Issuing for agents. At Skyfire, our illustrated checkout flow ties a mandate for a merchant, product, and amount to Agentic Wallet authorization of a tokenized card. Our wallet also supports managing stablecoins, but a merchant accepting a stablecoin payment still needs a compatible payment path and settlement arrangement. Identity and mandate checks do not replace either one.

We recommend setting merchant-side policies for per-order and cumulative spend caps, approval thresholds, and mandate scope by merchant. Your checkout can reject an order when its final amount exceeds the mandate, or pause it for human approval when a policy threshold requires it. Record the authorization decision before sending the payment request, and require a fresh decision if the cart or amount changes. These are recommended controls for your integration, not a claim that any one vendor provides every policy setting.

Handling fraud, declines, and disputes on agent-initiated orders

Agent-initiated orders need fraud checks that connect the payment to the agent’s authority to buy. Stripe says Radar assesses agent-initiated transactions and Shared Payment Tokens carry risk signals. Those signals can inform a fraud decision, but you should also check the order against the buyer’s mandate. A valid payment credential alone cannot show whether the buyer authorized that agent to purchase that item at that price.

We recommend treating a fraud decline, mandate mismatch, or request for additional authentication as a stop, not a cue to retry automatically. Send the order to a human for approval or correction, and preserve the decline reason. Reserve automated retries for identified processing failures when your payment provider permits them. Use an idempotency key so a repeated checkout request cannot create a second order.

Card disputes can surface well after an agent completes checkout, while fulfillment may begin immediately. Review risk signals before releasing the order, and retain the agent identity reference, mandate, checkout details, and payment decision for a later dispute. If a buyer disputes an agent order, pause any repeat purchase under the same mandate and ask a human to review the authorization before the agent transacts again.

Settlement and audit evidence

Card and stablecoin payments can settle on different timelines, but settlement alone does not show whether an agent had permission to place an order. A third-party comparison describes card settlement as taking multiple days. An x402 guidecites roughly two seconds for a USDC transfer on Base, with an on-chain fee below one-tenth of a cent. Those figures do not guarantee a merchant’s payout time or total cost. Your processor, conversion steps, and chosen rail can change both.

We recommend keeping an audit record that connects the agent identity reference, the user’s mandate scope, the authorization decision, and the payment and settlement references. Record when the decision occurred and retain the order details used to check the mandate. Skyfire’s illustrated checkout flow scopes a user mandate to a merchant, product, and amount, but merchants should verify what evidence each vendor actually supplies and how long it remains available. Treat this record as a requirement you set for vendors, not as a universal audit standard.

Controls matrix: ownership and failure handling

We recommend assigning an owner and a failure response to each control before accepting agent orders. This matrix is an operating model, not a claim that any single vendor provides every control.

Shopify vs BigCommerce vs Stripe vs Skyfire: which layer each one solves

Shopify, BigCommerce, and Stripe address merchant commerce or payment infrastructure. We position Skyfire as a complementary layer for checking which agent is transacting and whether its payment matches a delegated mandate. A merchant must confirm that its checkout accepts the credentials an agent presents.

Choose Shopify when you want its storefront and UCP path, or BigCommerce when you plan to build on catalog and checkout APIs. Choose Stripe when its US-only Suite fits your payment-acceptance needs. We recommend evaluating Skyfire alongside those providers when you need to identify the agent and check its delegated spending authority. Verify credential acceptance and integration details before committing to that design.

FAQs

Does a merchant need to choose one protocol? No. ACP, UCP, AP2, and x402 serve different transaction flows, so we recommend choosing integrations based on where your customers buy and what your checkout provider supports. Treat support for each protocol as a vendor-specific question. Do not assume that adding one lets you accept transactions through the others.

Does verifying an agent satisfy KYC or other compliance obligations? No. Agent identity controls can help you establish which agent acted, whose authority it used, and what evidence to retain. KYC requirements concern the people or businesses behind a transaction. Your compliance team must assess the obligations that apply to your business and payment methods.

Can existing OAuth or IAM handle agent checkout? Existing tools can authenticate services and grant scoped access, especially in bounded workflows. OAuth permission alone does not prove which autonomous agent initiated a purchase or whether its delegated authority covers the merchant, product, and amount. We recommend checking those details at authorization rather than treating login as purchase approval.

Does Skyfire replace the merchant’s payment processor? No. Skyfire describes KYA as an agent identity token and its Agentic Wallet as a way to collect user mandates and authorize payment credentials. Your payment processor still handles its part of the charge and settlement. Skyfire’s illustrated checkout ties a mandate to a merchant, product, and amount, but merchants should verify acceptance and integration requirements for their own checkout.

Where merchants should start

We recommend using the controls matrix to assign an owner and failure response to each checkout control. Use the comparison table to identify which parts your commerce platform and payment provider already handle, then assess whether you need an agent identity and delegated-authorization layer. Agent checkout should extend your existing commerce infrastructure, not replace it.

Merchants already receiving agent-initiated orders should act now, especially if they cannot identify the acting agent or verify its authority before charging a customer. Merchants without agent traffic can wait on an integration, but should ask their platform and payment providers which agent checkout flows they currently support. Skyfire can help evaluate agent identity and payment authorization. Merchant acceptance still depends on the credentials and checkout flows each integration supports.