Your Agent Can Now Pay at Checkout Without Ever Seeing Your Card

Checkout is the one part of the user journey most agents can't finish. They can browse a site, fill a form, and get all the way to the payment page, but then stop because nobody wants to give a real card number to an AI model.

Agentic Checkout with Stripe Link closes that last step: your agent completes the purchase, you approve the charge, and your card details never enter the loop.

What is Agentic Checkout with Stripe Link?

Agentic Checkout with Stripe Link enables an AI agent to complete a real purchase on a merchant's own checkout page, paying from your own Stripe Link wallet instead of a card handed to the agent.

It ships as two MCP tools and a wallet card on the Integrations page of your dashboard. browserless_link_connect manages the wallet connection, and browserless_link_checkout creates the spend request you approve and fills the approved card into the merchant's payment form. Your agent still uses browserless_agent to reach the payment page and submit the order.

You stay in control of your money the whole way. Every purchase needs your approval in Link before any money moves, Browserless never holds your funds or keeps a card on file, and each card works for one approved purchase only.

Agentic Checkout is available to every Cloud account that uses the full Browserless MCP server. It works with Stripe Link wallets registered in the US, for checkouts charged in US dollars, up to $50.00 a purchase during the beta, and splitting an order to get under that cap isn't permitted. It isn't available on Dedicated or self-hosted deployments, so contact us if you need it on yours.

Why we built Agentic Checkout

The point of an agent is to finish a task for you, and a lot of tasks end at a checkout page. Stripe Link's support for agent payments is what makes this possible without Browserless ever holding your money or your card: you authorize the spend, Stripe issues the one-time card, and Browserless is the browser that types it in.

That's also why we built this as a wallet connection rather than a stored payment method. A stored card sitting on a server is a target. A wallet you approve per purchase, with a card that only works once, gives your agent exactly the access it needs to finish the job and nothing more.

How to get started

1. An account owner or admin connects your Stripe Link wallet once from the Integrations page in the dashboard.
2. Point your MCP client at the full Browserless MCP server. The Connector and compliance-mode endpoints don't include the checkout tools.
3. Ask your agent to buy something. It goes to checkout, requests the spend, and you approve it in Link.

The Agentic Checkout docs cover the full setup and limits.

How Agentic Checkout works

A checkout runs inside one browserless_agent session and uses two tools: browserless_link_connect to check the wallet, and browserless_link_checkout to create the spend request, fill the card, and report the result.

1. Check the wallet. The agent calls browserless_link_connect with status to confirm your Stripe Link wallet is connected.
2. Reach the payment page. browserless_agent browses to the merchant's checkout and reads the cart and payment fields from the page.
3. Create a spend request. At the payment step, browserless_link_checkout with action: "create" sends the merchant, the cart, and the exact total in US dollars. It returns an approval link on a Stripe-owned domain.
4. Approve in Link. You open the link and approve that amount, and Link issues a single-use card scoped to that purchase.
5. Fill the card. Within 10 minutes of the request, action: "resume" fills the approved card into the merchant's payment fields and returns filled, plus the optional last four digits. If Link needs an extra step, such as 3D Secure, the checkout returns requires_action. Depending on the step, the same checkout resumes once you complete it, or the agent starts a new spend request.
6. Submit and report. The agent submits the order, checks the merchant's confirmation page, and records the outcome with action: "report".

The full card number, CVC, and expiry never appear in a tool result or reach the model. Each checkout is also bound to the session and page that created it, so a resume fails if the page origin or payment fields change in between.

Get more out of Agentic Checkout

Agentic Checkout runs in the same Browserless session as the rest of your agent's work, so the features it already uses help it reach checkout more often.

Reach more checkouts

Sessions can run with stealth fingerprints, CAPTCHA solving, and residential proxies. Together, they get your agent past a lot of the bot protection that stops automated checkouts before the payment step. A well-protected merchant can still block a session, but if that happens before the spend request, nothing is charged.

Check out behind a login

If a store needs an account, the agent signs in with a saved Authenticated Profile, then continues to checkout in the same run. Profiles keep a merchant login reusable across sessions, so you don't have to paste a password into the prompt.

Roll it out to your team

An owner or admin connects one Stripe Link wallet, and anyone on the account with an API key can run checkouts through it. Every charge still needs the buyer's approval in Link, so treat API keys as secrets.

What's next

Browsing, form-filling, and logging in were already part of the agentic commerce flow. What was missing was a way for an agent to pay without ever gaining access to sensitive financial information. Agentic Checkout with Stripe Link fills that gap, and it's available on every Cloud account now.

The best way to see it is to start small. Connect your Stripe Link wallet, then have your agent buy something low-cost from a merchant you already use, so you can see the full approval flow before relying on it.

Agentic Checkout with Stripe Link FAQs

Does my agent ever see my card number?

No. Stripe issues the one-time card, and Browserless fills it into the merchant's payment form server-side, inside the session. The card values never appear in a tool result and are redacted from your agent's page snapshots, so the only card detail that can come back to your agent is the optional last four digits.

Who approves each purchase?

You do, every time. Your agent requests a spend, and you approve it in Stripe Link before the one-time card is issued. Browserless and your agent never gain the ability to charge your wallet on their own.

Can I use Agentic Checkout from Claude Code or Claude.ai?

Yes, as long as the client connects to the full Browserless MCP server. Claude Code does this with claude mcp add, and Claude.ai does it with a custom connector that points to the full server URL. The Browserless Connector endpoint and compliance mode don't expose the checkout tools.

What happens if my agent reaches a merchant that blocks it?

If the block happens before the payment step, no spend request is created, and nothing is charged. If the merchant blocks or declines the payment after you've approved a spend request, your agent reports the checkout as blocked. Check the request in Link and cancel it there if it's still pending. The single-use card only covers that one approved amount and stops working after the 10-minute window.

Does anything change if I don't use Agentic Checkout?

No. The two MCP tools and the wallet card are additive, so your existing sessions, scripts, and integrations behave exactly as they did before. Nothing happens until an owner or admin connects a wallet.