Stripe and Tempo just introduced the Machine Payments Protocol (MPP), an open, production-grade standard designed for agent-to-agent (A2A) transactions. It reframes payments not as a separate step, but as something embedded directly inside software workflows.
What it does
At its core, MPP defines a simple but powerful flow: an agent requests a resource, receives a payment requirement, authorizes it, and gets delivery. No checkout pages, no redirects—just a native interaction loop between services. This is a shift away from human-centered UX toward machine-native execution.
The real unlock is the session primitive. Instead of approving every transaction individually, an agent can be authorized with a budget upfront and stream payments as it works. That eliminates per-transaction friction and enables real-time, usage-based commerce across APIs, data, compute, and services. MPP is also payment-method agnostic, supporting cards, stablecoins, and other rails, which makes it flexible enough to operate across both traditional and crypto ecosystems.
Partner integration with Visa
Visa extends this model with its CLI, adding a critical identity and trust layer. Agents can execute authenticated, tokenized card payments directly from the command line, with consent and authorization embedded into each transaction. This connects existing global card infrastructure with emerging agent-driven systems, making the model viable at scale.
Taken together, this begins to look like a foundational layer for agentic commerce. Agents are no longer just generating outputs—they can now transact, coordinate services, and participate economically in real time.
What still needs to be built
But an important piece is still missing.
MPP answers the question: can an agent execute a payment?
It does not answer: should it?
That gap is the intent governance layer—how spend limits, vendor rules, permissions, and policies are enforced deterministically. Deciding whether an agent should spend money in a given context requires systems that go beyond protocols: policy engines, approval frameworks, and control layers that sit between intent and execution.
Permission and intent are not the same. Standardizing payment flows is necessary, but not sufficient.
The next wave of infrastructure in agent-driven commerce won’t just move money—it will define how decisions to spend are made.