Stripe and Tempo launched the Machine Payments Protocol (MPP) today—March 18, 2026—an open standard that enables AI agents to make autonomous payments without human intervention. The protocol solves a fundamental infrastructure problem: traditional payment systems require manual account setup, form filling, and billing configuration that AI agents can’t provide. Real-world implementations are already live. Browserbase uses MPP for per-session browser payments, PostalForm for printing physical mail, and Prospect Butcher Co. for ordering sandwiches in NYC. Design partners include Anthropic, OpenAI, Visa, Mastercard, and DoorDash.
As AI agents become more autonomous, they need payment capabilities to access APIs, data, and services programmatically. MPP removes the infrastructure barrier that’s been forcing developers to either hardcode payment credentials (insecure), require human approval for every transaction (defeats automation), or avoid paid services entirely (limits functionality).
Traditional Payment Systems Weren’t Built for Bots
Here’s the core problem: every payment system you’ve ever used was designed for humans. Creating an account requires filling out forms. Making a purchase requires navigating checkout flows, entering payment details, and often solving CAPTCHAs specifically designed to block automated software. AI agents can’t do any of this.
According to Stripe’s Engineering Lead Steve Kaliski, “Traditional payment systems require human intervention—account creation, navigation, payment details entry. MPP eliminates these friction points for autonomous operations.” The impact is already visible. Browserbase, valued at $300 million after raising $40 million in June 2025, now enables AI agents to spin up headless browsers and pay per session without any manual setup. That wasn’t possible before MPP.
How the Machine Payments Protocol Works
MPP uses a straightforward request-response flow. An agent requests a paid resource from a service. The service returns HTTP 402 (“Payment Required”) with payment details. The agent authorizes payment within pre-set spending limits. The service delivers the resource. Done.
The protocol introduces a “sessions” primitive—think of it as OAuth for money. Agents authorize spending limits upfront, then payments execute automatically within those parameters. Critically, sessions aggregate thousands of small transactions into a single settlement. This makes true micropayment economics viable at internet scale. An agent could make hundreds of API calls with micropayments for each call, all settled as one batch.
For developers, Stripe integration uses the existing PaymentIntents API. According to official documentation, you can accept MPP payments “in just a few lines of code.” The underlying infrastructure is serious. Tempo blockchain, which co-launched today, promises 100,000+ transactions per second with sub-second finality. That vastly outperforms Ethereum (~20 TPS) and even Solana (few thousand TPS in practice).
Real Companies, Real Payments, Today
Three production implementations demonstrate MPP’s range. Browserbase handles per-session headless browser payments—critical infrastructure for AI agents that need web browsing capabilities. PostalForm enables agent-initiated physical mail printing. An agent can autonomously send contracts, invoices, or legal documents via postal mail. Prospect Butcher Co. lets agents order sandwiches for human pickup or delivery anywhere in NYC.
That last example matters more than it sounds. MPP isn’t limited to digital services. Agents can interact with real-world services through automated payments. The protocol’s design partners reinforce this breadth. Visa extended MPP for card payments. Lightspark added Bitcoin Lightning support. Major tech companies including Anthropic, OpenAI, DoorDash, Mastercard, Nubank, Ramp, Revolut, Shopify, and Standard Chartered are involved.
Crypto or Fiat? Both.
MPP is rail-agnostic by design, supporting both cryptocurrency payments (stablecoins on Tempo blockchain) and traditional fiat (credit cards, buy-now-pay-later via Stripe). Stripe acts as orchestrator, routing payments through optimal rails—card networks, ACH, or Tempo—based on cost and speed. Developers see a unified API regardless of payment method.
This matters because many developers are crypto-skeptical or work for companies that can’t touch blockchain tech. The multi-rail approach means you can use MPP with traditional Stripe payments (cards, bank transfers) and optionally add stablecoin support later. Shared Payment Tokens (SPTs) enable fiat payments through Stripe’s existing infrastructure, complete with fraud protection, tax calculation, and accounting integrations. No blockchain required if you don’t want it.
The Regulatory Problem Nobody’s Solved
MPP faces significant regulatory challenges, particularly in EU/UK markets where Payment Services Directive 2 (PSD2) and Strong Customer Authentication (SCA) require clear human authorization for payments. Currently, there’s no legal mechanism for AI agents to be treated as equivalent to human payers. All SCA authentication methods are tied to a person—the payer.
Legal analysis from Pinsent Masons notes that “as these systems become more autonomous, it will become more difficult for PSPs to demonstrate a transaction was appropriately authorised.” The liability question is critical: if an agent makes an unauthorized payment due to AI hallucination or bug, who’s responsible? The customer? The service provider? The agent developer? Consumer trust reflects this uncertainty—only 16% of US consumers currently trust AI to make payments.
Security experts warn that agents holding wallets create a new attack surface. Compromised agents could drain funds. Developers building on MPP need spending limits, audit logging, and potentially insurance to protect against agent malfunctions. This isn’t just a technical consideration—it’s a business and legal risk that existing regulations don’t adequately address.
What This Means for Developers
MPP opens new possibilities for autonomous agent workflows, but the ecosystem is nascent. Real production implementations exist (Browserbase, PostalForm), and major partners (Anthropic, OpenAI, Visa, Mastercard) signal industry-wide support. However, regulatory frameworks lag significantly behind the technology. EU/UK compliance remains unclear, and liability questions are unresolved.
For developers building AI agents, MPP removes a fundamental infrastructure barrier. Agents can now programmatically access paid services without manual intervention. The multi-rail approach (crypto + fiat) lowers adoption barriers—you’re not forced into blockchain tech if your business can’t support it. But implement spending controls, maintain audit trails, and understand the regulatory landscape of your target markets. The technology is ready. The legal framework isn’t.
