← cartapi.io blog

x402 vs Stripe Checkout: which payment rail should your agent use?

Published 2026-11-11 · Last updated 2026-11-11 · 8 min read · engineering
TL;DR

x402 and Stripe Checkout solve different problems. x402 lets software pay for an HTTP request in a stablecoin, with no account and no page to click through. Checkout lets a person pay by card on a hosted page. Stripe now supports x402 too, along with its own MPP protocol, which also takes cards. If the agent pays for its own API calls, use x402 or MPP. If a person is buying something, like a gift card, top-up or eSIM, use a hosted checkout they approve.

"x402 vs Stripe" sounds like a protocol war, and most of the pages that rank for it are written by companies selling one side. The useful question is narrower: who is paying, for what, and when do they agree to it? Answer that and the choice mostly makes itself. Here's how each rail works, a decision table, and what we chose for cartapi.

x402 in one paragraph

x402 is an open payment standard, licensed Apache-2.0, built on the HTTP 402 Payment Required status code. A client asks for a resource. The server replies 402 with payment requirements. The client builds a payment payload and retries. The server verifies and settles the payment itself or through a facilitator's /verify and /settle endpoints, then returns the resource. The docs stress that clients pay "without accounts, sessions, or credential management", and that the protocol has no fees of its own. The use cases it lists are per-request APIs, agents paying for API access, content paywalls and microservices.

GET /paid HTTP/1.1

HTTP/1.1 402 Payment Required
payment-required: eyJ4NDAyVmVyc2lvbiI6MiwiZXJyb3IiO...

That header is what Stripe's x402 guide shows coming back from a test server. It's a base64-encoded payload saying what to pay, on which network and to which address.

Stripe Checkout in one paragraph

Checkout is a Stripe-hosted payment page. Your server creates a Checkout Session for a specific order and gets back a URL. The person opens it, pays by card or wallet, and your webhook fulfils the order. In an agent, the MCP tool returns that URL and the agent hands it to the user. The agent never touches the card. We walked through the whole build, including the webhook and manual capture, in How to add Stripe Checkout to an MCP server.

Where Stripe's machine payments fit

The "vs" got blurrier in 2026. Stripe's machine payments docs cover two 402-style protocols:

Availability is the practical constraint. Stripe's stablecoin payments are open to businesses in every US state except New York. Businesses elsewhere email Stripe to request access in 30+ countries, and the payment method only goes live after Stripe reviews the account. Payments land in your Stripe balance and settle in fiat, and Stripe says refunds work through the Refunds API.

The comparison

x402Stripe MPPStripe Checkout
Who paysSoftware with a walletSoftware with a wallet or a Shared Payment TokenA person
When consent happensBefore: whoever funded the wallet set its limitsBefore: the token or wallet carries the limitsPer purchase, on the payment page
Payment methodsStablecoins (via Stripe: USDC on Base)Cards via SPTs; USDC on Tempo and SolanaCards, wallets and local methods
MinimumProtocol: none. Via Stripe: 0.01 USDC0.50 USD (cards), 0.01 USDC (stablecoins)Stripe card minimums (0.50 USD)
Buyer account neededNoNo checkout UI; an SPT comes from a wallet such as Link'sNo account, but a human at a browser
RefundsNo protocol step. Via Stripe: Refunds APIRefunds APIRefunds API; card disputes apply
Card fraud tools (3DS, issuer country)Not applicableCard rails, but no page for the cardholderYes
Best fitPer-call APIs, data, content, sub-cent pricingPer-call or session pricing that should also take cardsA person buying a product

Read the "when consent happens" row first. With x402 and MPP, consent is given up front, in whatever limits the wallet or token carries. With Checkout, it's given on every purchase. Neither is better in general. They put the risk in different places, which is the same trade-off we covered for ACP envelopes in Consent-first checkout.

A build-vs-buy checklist

Answer these about your own product before you pick a rail:

  1. Is the payer the agent or a person? If the agent is paying its own running costs (search calls, data, inference), that's x402 or MPP territory. If it's buying something on a person's behalf, start from Checkout.
  2. What's the price? Cents or less per call rules out card checkout. A $25 gift card doesn't need micropayments.
  3. Does fraud control depend on card data? If you need 3DS, the card's issuing country, or a manual-capture review, you need a card payment the cardholder completes.
  4. Who handles refunds and disputes? Consumer goods produce refund requests. Make sure your rail and your merchant of record can handle them. See Merchant of record for AI-agent purchases.
  5. Can you get access? Stablecoin acceptance through Stripe depends on where your business is based and on Stripe's review. Check before you design around it.
  6. Do your buyers' agents have wallets? x402 only helps if the paying client holds a stablecoin wallet. If your buyer is a person chatting with a consumer assistant that has no wallet, a link is what they can act on.

What we chose for cartapi, and why

cartapi sells digital goods that people use: gift cards, mobile top-ups and eSIMs. Every purchase goes through a Stripe-hosted Checkout link that the person opens and pays. Three reasons that aren't about taste:

Where x402 or MPP could make sense for us is the other side of the API: an agent paying per call for catalog search, without an API key. That's an idea, not a roadmap item. Today, REST calls use a bearer API key and the MCP endpoint is open.

Combining them

These aren't mutually exclusive. One service could use MPP or x402 for metered API access and hand back a Checkout link for anything a person buys. Stripe's docs point that way too: they recommend adding MPP next to x402 if you also want to accept cards. If you're mapping the wider protocol stack (MCP for tools, ACP for checkout, A2A between agents), start with ACP vs MCP vs A2A.

Frequently asked questions

What is the difference between x402 and Stripe Checkout?

x402 lets software pay for an HTTP request in a stablecoin after a 402 response, with no account or page. Checkout is a hosted page where a person pays by card or wallet.

Does Stripe support x402?

Yes, for USDC on Base. Payments are recorded as PaymentIntents using a preview API version. Stablecoin acceptance is open to US businesses outside New York; others request access.

What is Stripe MPP, and how is it different from x402?

MPP, co-authored by Stripe and Tempo, also uses a 402 challenge, but accepts cards through Shared Payment Tokens (from 0.50 USD) as well as stablecoins (from 0.01 USDC).

Should an MCP server use x402 for purchases?

If the agent pays for its own calls in cents, x402 or MPP fits. If a person is buying a product, return a hosted checkout link they approve.

Can x402 payments be refunded?

The protocol has no refund step, and a stablecoin transfer has no card chargeback. Through Stripe, refunds go through the Refunds API.

Or skip the build

If your agent needs to sell gift cards, top-ups or eSIMs to people, cartapi already does the checkout, fraud gates and fulfilment. Add it from the install page, or read the docs for the REST flow.

About the author

Roc Chow

Roc Chow builds cartapi, the commerce layer for AI agents, and runs . The x402, MPP and Stripe details here come from the x402 docs and Stripe's machine payments docs, linked above, as read on the day of writing. The cartapi checkout details come from its own codebase.

LinkedIn · · cartapi.io