Mastercard opens early access for agents buying digital services within an approved budget

An Agent Pay for Machines demonstration uses signed payment vouchers and x402 to connect research and image-generation services in one task.

Mastercard has opened early access to Agent Pay for Machines, giving developers buyer- and seller-side APIs for agents that purchase digital services within a defined task and budget. The company demonstrated the workflow at Cloudflare Connect on 20 October, following end-to-end pilots with partners across payments, wallets and AI services.

The system builds on x402, an open protocol through which an online service can request payment. Mastercard adds a common trust layer intended to establish who authorised an agent, what it may buy and how much it may spend.

Paying as the task develops

The workflow begins with a business assigning a job and a spending limit. When the agent reaches a paid service, the provider returns an HTTP 402 response. The agent creates a digitally signed payment voucher within its approved authority, and the provider verifies it before delivering the service.

The interactions occur off-chain. Providers can aggregate vouchers and redeem them later, separating the immediate service exchange from settlement. This can let a task use several providers without requiring a person to establish an independent billing relationship for every individual service along the way.

That is particularly relevant to agents whose next tool depends on what they discover. A research task may need a specialised dataset, followed by a model call and a compute job. A common payment mechanism can make those choices more dynamic while keeping them inside the business's original spending scope.

A campaign pilot with two merchants

Mastercard's demonstration uses marketing agency Hype and a sample protein-bar campaign. The agent proposes research from AIsa and campaign imagery through Cloudflare AI Gateway. Hype approves the funding, after which the agent purchases the research and eight images.

The demonstration therefore comprises nine payments to two merchants within one approved budget. It is a pilot example of the payment sequence, rather than a claim about the sales performance of the generated campaign.

The described infrastructure includes Coinbase Developer Platform for funding-wallet infrastructure and Base for on-chain settlement. BVNK and Mastercard's Global Money Platform handle payment and settlement arrangements, with sellers able to receive their preferred supported currency, token or rail.

Two sides of an agent market

Early access provides the first buyer- and seller-side APIs, a toolkit, sandbox and developer documentation. That gives builders two distinct integration tasks: enabling an agent to recognise and pay an authorised request, and enabling a service to verify the payment promise before supplying its output.

The seller's part is important. A provider needs a machine-readable way to describe the required payment and a reliable rule for accepting the voucher. The buyer needs enough information to decide whether the purchase belongs to the task and fits the remaining budget.

For businesses, the unit of authority is the approved assignment. A budget alone does not describe the intended purchase; the task and permissions give the spending its context. Mastercard's design places those conditions in the payment workflow instead of relying on a conversational instruction alone.

The announcement opens a developer programme rather than declaring unrestricted general availability. Its concrete contribution is an early-access path for testing how agents can buy models, data, APIs and computing services as part of a continuing job, with payment verification and settlement carried by a shared infrastructure.

Sources

- [What happens when AI agents start paying each other? Watch Mastercard Agent Pay for Machines in action](https://www.mastercard.com/us/en/news-and-trends/press/2026/october/Mastercard-Agent-Pay-for-Machines-in-action.html) — Mastercard