What is Stripe agentic commerce?
It is the umbrella Stripe uses for two different problems that happen to share a payment rail. One is a selling problem: you run a business, shoppers are asking AI assistants for products, and you want your catalog and your checkout to be reachable from inside those conversations. The other is a buying problem in miniature: your API or service is worth money, an agent wants to call it, and there is no human present to create an account and type in a card. Stripe’s documentation keeps these apart deliberately, and it is worth keeping them apart in your own planning too, because the integration work, the protocols and even the economics are different. Almost every article that treats "Stripe agentic commerce" as one thing is describing only the first track.
What is in the Stripe Agentic Commerce Suite?
Stripe announced the Agentic Commerce Suite on December 11, 2025 as a way to collapse what it described as an integration that "can take up to six months for every new AI agent you support" into a single one. Four components: a hosted ACP endpoint that publishes product, price and availability data to agents in near real time; a Checkout Sessions API that handles the parts of a purchase merchants underestimate, such as shipping calculation and tax; Shared Payment Tokens as the payment primitive; and Radar, extended with fraud signals specific to agentic traffic including the likelihood of a fraudulent dispute, card testing, a stolen card or an issuer decline. Stripe says the components are "modular by default", which in practice means you can take the payment primitive without the hosted feed, or the feed without the checkout. Launch names included URBN with Anthropologie, Free People and Urban Outfitters, plus Etsy, Ashley Furniture, Coach, Kate Spade, Nectar, Revolve, Halara and Abt Electronics, alongside platform partners Wix, WooCommerce, BigCommerce, Squarespace, commercetools, Akeneo, Cymbio, Logicbroker, Mirakl, Pipe17 and Rithum.
Does Stripe use ACP or UCP?
Both, and this is the single most out-of-date thing in the existing coverage. Stripe co-authored the Agentic Commerce Protocol with OpenAI and released it under Apache 2.0 in September 2025, so for a year the assumption was that Stripe meant ACP whenever it said agentic commerce. As of August 2026 Stripe’s own docs present the seller integration with the protocol column reading "UCP or ACP", listing Google’s Universal Commerce Protocol first. Stripe has positioned itself as rail-neutral between the two competing standards rather than as ACP’s house processor. If you are a merchant choosing a protocol, that is genuinely useful: it means the payment side of your integration does not have to be the thing that decides the protocol question for you, and you are not stranded if the standard you picked loses.
What are Stripe shared payment tokens?
A shared payment token, or SPT, is a scoped grant that lets an agent charge a customer’s saved payment method without ever seeing the underlying credential. Each token is bound to a specific seller, a maximum amount and an expiry window, and its lifecycle is observable, so a token that leaks is worth nothing outside the merchant and ceiling it was minted for. SPTs work with ordinary PaymentIntents, which is the reason adoption moved quickly among businesses already on Stripe. They are available to businesses in all US states, can be denominated in any Stripe-supported currency, and are issued from the Link Agent Wallet. On March 3, 2026 Stripe extended SPT support to the card networks’ own agentic credentials, Mastercard Agent Pay and Visa Intelligent Commerce, and to buy now pay later through Affirm and Klarna, noting that any seller already processing with Stripe automatically supports the new methods for agentic transactions.
What is the Machine Payments Protocol (MPP)?
MPP is Stripe’s protocol for the second track, the one where an agent pays for an API call rather than a sweater. The shape is familiar to anyone who has looked at HTTP 402 designs: your server returns a payment challenge, the agent presents a valid payment credential, and Stripe settles into your balance. What makes it commercially interesting is that MPP accepts both cards and stablecoins, which is unusual. Most machine-payment schemes are stablecoin-only, and that rules out every seller who wants fiat in a bank account without touching crypto operations. Money lands in your Stripe balance and settles in fiat, refunds run through the ordinary Refunds API, and reporting and multi-currency payouts behave like any other Stripe payment. Machine payments are also available to Connect platforms across all charge types, so a marketplace can pass agent revenue through to its sellers.
Does Stripe support x402?
Yes, for stablecoins only. Stripe documents x402 alongside MPP as a supported way to accept machine payments, and the split between them is concrete rather than philosophical. MPP runs stablecoins on Tempo with USDC.e and on Solana with USDC; x402 runs on Base with USDC. Card acceptance is MPP only. There is a jurisdictional wrinkle worth knowing before you build: shared payment tokens are available to businesses in all US states, but stablecoin acceptance is available in all US states except New York, and businesses outside the US have to email Stripe with an account ID to request access. If your seller base includes New York entities, the card path through MPP is the one that works everywhere.
What is the Stripe Agent Toolkit?
The Agent Toolkit is Stripe’s open-source library for wiring Stripe API calls into agent frameworks as callable tools. It ships for Python and TypeScript, sits directly on top of the existing Stripe SDKs, and supports OpenAI’s Agents SDK, LangChain, CrewAI, Vercel’s AI SDK and the Model Context Protocol. Its home has since broadened into the stripe/ai repository, presented as a one-stop shop for building AI products on Stripe, and the toolchain has grown around it: the Stripe CLI from version 1.43.3 supports stripe agent setup, and Stripe publishes agent skills and plugins so a coding agent can integrate Stripe correctly without a human reading the docs first. Treat the toolkit as capability plumbing. It is what lets an agent call Stripe. It is explicitly not what decides whether the agent should have. How far that capability reaches, and how Stripe's generic stripe_api_write tool compares with the named tool lists PayPal and Square publish, is the subject of our probe of payment MCP servers.
Can AI agents pay with Stripe today?
On the selling side, yes and at scale. On the buying side, partly. The seller-facing pieces are generally available: SPTs, MPP, x402, the hosted ACP endpoint, Radar for agentic traffic. The track aimed at companies building the agent itself, where you embed a storefront and a checkout into your own AI interface, is still in private preview behind a waitlist as of August 2026. Funding an agent is the part that does work now, through the Link Agent Wallet for cards or a crypto wallet for stablecoins. That asymmetry is a reasonable read of where the demand actually is: far more companies want to be found and bought from by agents than want to run a consumer shopping assistant of their own.
What does Stripe agentic commerce cost?
Stripe has published no separate rate card for any of it, and it is worth being precise about what that does and does not mean. The Agentic Commerce Suite announcement disclosed no fees. The machine payments documentation discloses no surcharge. What Stripe does publish is minimums rather than prices: 0.50 USD is the floor for a card payment through a shared payment token, and 0.01 USDC for a stablecoin payment. Those two numbers matter more than they look, because they define whether per-call metering is viable for you at all. If your API is worth a tenth of a cent per call, the card path cannot price it and you either batch into larger settlements or go the stablecoin route. The other cost people conflate with this is the platform take rate on AI shopping surfaces, reported in a 2% to 4% band for ChatGPT Instant Checkout and never confirmed at source. That is a channel fee, not a Stripe fee, and it should sit in a different line of your model.
What Stripe agentic commerce does not do
Everything above is per-transaction, and the omission is structural rather than an oversight. A shared payment token caps one grant to one seller for one amount until one expiry. MPP settles one call. An x402 challenge prices one request. Not one of those primitives knows that the agent holding the token already burned four thousand dollars this month across nine merchants and two rails, because nothing in the design is keeping a running total. This is not a criticism unique to Stripe, it is the same gap in Google’s AP2, in ACP and in x402, and it exists because a payment rail is scoped to authorize transactions rather than to govern budgets. The clearest evidence that it is a real gap rather than a talking point is that Stripe’s own Agent Toolkit repository carries an open issue, filed April 6, 2026, asking Stripe for a governance layer covering commerce delegation, spend limits, merchant allowlists and an audit trail, specifically requesting per-task spending ceilings beyond what an API key scope can express, merchant allowlists, human approval gates above a threshold, and an audit trail that traces a payment back to the human who delegated the authority.
Where a control plane fits alongside Stripe
Nothing here is a reason not to use Stripe. If you sell, the Agentic Commerce Suite is the shortest path onto AI surfaces, and if you meter an API, MPP is the only mainstream option that takes cards as well as stablecoins. The layer that is missing sits above all of it. Agentspay gives each agent its own wallet with a hard limit, so an agent is structurally unable to spend past its budget regardless of how many tokens it was granted or which rail it used. Purchases above a threshold you set pause for human approval before the money moves rather than after. Scoped virtual cards constrain an agent to named merchants and ceilings on rails that never touch a shared payment token at all. And every decision, the policy that allowed it, the agent that asked and the person who approved it, lands in one immutable audit trail that covers card spend, stablecoin settlement and metered API calls in the same place. Stripe authorizes the transaction. The control plane decides whether it should have happened, and can answer for it afterwards.
How Stripe compares to the other agent payment stacks
Stripe is the broadest of them on rails and the most conventional on economics, which for most US businesses is the right combination. Money settles in fiat into an account you already reconcile, refunds and disputes work the way your finance team already knows, and Radar brings a decade of fraud modeling to a traffic pattern that is barely a year old. Compared with a stablecoin-native stack you give up some settlement speed and gain a treasury process nobody has to invent. Compared with the card networks’ own programs you get one integration instead of two, since Stripe now brokers both Agent Pay and Intelligent Commerce through the same token. Where you will still need something else is anywhere the question is organizational rather than transactional: which agents exist, what each is allowed to spend, who signs off on exceptions, and how you prove it later. Our comparison of AI agent payment platforms lays out where each vendor sits on that line.
What to watch next
Three things. Whether Stripe’s neutrality between UCP and ACP holds or whether one protocol’s volume forces a default, because merchants will not maintain two feeds forever. Whether the agent-side private preview opens generally, which would tell you how seriously Stripe rates the market for companies building their own shopping assistants rather than selling into someone else’s. And whether issue 356 ever ships, because if Stripe does grow a governance layer it will define the vocabulary everyone else has to speak, and if it does not, that confirms the layer belongs above the rail rather than inside it. We keep this page current as those move.