Machine Payments Protocol (MPP): How AI Agents Pay for Services

The first time an AI agent tries to do something useful on the internet, it hits a wall. It wants to call a paid API, access a dataset, rent compute time, or pay for a model inference. There is no checkout form to fill in, no card to tap, and no human in the loop to confirm. The agent simply cannot pay.

That is the problem the Machine Payments Protocol is designed to fix.

MPP is an open standard for machine-to-machine payments co-authored by Stripe and the Paradigm-backed blockchain startup Tempo. It launched on March 18, 2026, alongside the Tempo mainnet and provides a standardized way for AI agents and automated software to request, authorize, and settle payments without requiring human approval at every step.

This guide covers what MPP is, how it works technically, how it compares to competing agentic payment standards, and what kind of underlying stablecoin infrastructure is required for it to work reliably at scale.

MPP defines a shared protocol layer that sits between an AI agent and any service that wants to charge for access. Instead of each API or platform inventing its own billing logic, MPP establishes a common grammar: how a service announces its price, how an agent authorizes a payment, and how settlement is confirmed.

At its core, MPP lets an agent request a resource from a service, after which the service responds with a payment request. The agent then authorizes payment from its wallet, the transaction settles instantly, and the service delivers the requested resource.

The protocol was built to be payment-rail-agnostic. While MPP currently runs on the Tempo blockchain, Fortune reported that it is designed to operate across multiple blockchains and payment rails, not just one. Visa has already extended it to support card-based payments, Stripe has extended it for cards, wallets, and other payment methods through their platform, and Lightspark has extended it for Bitcoin payments over the Lightning network.

The broader promise is interoperability: an agent should be able to transact with any MPP-compatible service regardless of which chain the service settles on.

Understanding why MPP matters requires understanding the gap it fills. Most existing payment infrastructure assumes a human is present. A cardholder authenticates, a browser session redirects to a hosted payment page, and a person clicks confirm. These flows were never designed for software operating autonomously.

AI systems are evolving from assistants that respond to prompts into agents that take action. They book travel, manage subscriptions, and increasingly, spend money. The familiar model of redirecting to a hosted payment page simply does not work for agentic commerce at scale.

The problem becomes more acute when you consider transaction volume. A single AI-driven research workflow might call dozens of external APIs to complete a task. Each call needs to be compensated. Traditional payment rails, such as card networks, were designed for human commerce — not thousands of sub-cent payments between software services.

This is where stablecoins become important. Stablecoins offer programmability and composability, and the blockchains on which they run serve as a common reference point for settlement. They also settle in near-real time, operate continuously without banking hours, and support conditional logic through smart contracts. For agentic systems that need to transact at API speed, stablecoins' programmable properties map far more naturally to software execution than card networks do.

The market stakes are significant. McKinsey estimates US agentic commerce could see up to $1 trillion in orchestrated B2C revenue by 2030, with global projections ranging between $3 trillion and $5 trillion.

Understanding why networked stablecoins create new velocity at scale helps frame why programmable money is the right settlement layer for this shift. Liquidity that remains siloed across isolated blockchains creates drag at every step. A networked, programmable money layer removes that drag and makes the economics of agentic commerce viable.

The most technically interesting aspect of MPP is a primitive called sessions. Sessions allow for continuous payments and function something like OAuth for money: authorize once, then allow payments to execute programmatically within defined limits. When an agent opens a session, it sets aside funds upfront. As the agent consumes resources from services — an API call, a model inference, a data query — payments stream continuously without requiring a separate onchain transaction for each interaction. Thousands of small transactions can be aggregated into a single settlement transaction, making true pay-per-use payments viable at internet scale.

This design avoids the gas inefficiency of recording every micro-interaction onchain while maintaining a trustworthy settlement record. It is the same logic behind payment channels in Lightning but applied more broadly to any stablecoin-denominated service interaction.

MPP also ships with a discovery layer. The payments directory provides a unified catalog of MPP-compatible services that any agent can transact with automatically. At launch, it already includes integrations with more than 100 services spanning model providers, developer infrastructure, compute platforms, and data services — including Alchemy, Dune Analytics, Merit Systems, and Parallel Web Systems.

MPP enables a variety of use cases: pay-per-call APIs, monetized MCP servers, gated content and data, and multi-service workflows. The directory matters because agent discovery is as important as agent payment. An agent that can pay but cannot find what to pay for is incomplete. MPP combines the payment protocol with a registry of services that have already implemented it, reducing the integration burden for both sides of the market.

MPP is not the only protocol competing to become the standard for machine-to-machine payments. Two other significant frameworks have launched in parallel, and understanding how they differ clarifies what MPP is actually solving.

Coinbase's x402 protocol takes a different approach. It revives the HTTP 402 status code, "Payment Required," to embed stablecoin micropayments directly into the HTTP request-response cycle. The idea is that any web resource can respond to a request with a 402, specifying the payment amount and recipient, and a compatible client includes payment in the next request header. It operates at the protocol level rather than the application level, which makes it lightweight but also limited in expressing complex authorization flows.

Google's Agent Payments Protocol (AP2) builds trust through cryptographically-signed digital contracts called Mandates that serve as verifiable proof of a user's instructions. These mandates address both real-time purchases where a human is present and coordinated tasks where an agent acts on pre-authorized instructions. AP2 is also rail-agnostic, with support for cards, real-time bank transfers, and stablecoins through the x402 extension.

The three protocols are better understood as complementary layers in an emerging stack rather than direct competitors. x402 operates at the execution layer, enabling instant programmatic payments at the HTTP level. AP2 defines the trust and authorization model. MPP focuses on the coordination layer between agents and services, particularly for continuous streaming payments.

Each protocol reflects different priorities. x402 is lightweight and easy to implement. AP2 emphasizes verifiable authorization chains. MPP prioritizes continuity and the session model that makes streaming pay-per-use commerce practical without constant on-chain overhead.

The open question is fragmentation. There are multiple ways for agents to pay each other, but if they cannot all agree on how payments should work, it is difficult to bootstrap marketplaces — whether they use micropayments or not. The industry is still in early standardization, and which protocol gains dominant adoption will likely depend on which has the most developer-friendly tooling and the deepest service directory.

A protocol standard is only as useful as the infrastructure running beneath it. MPP can define how agents negotiate and authorize payments, but it cannot by itself ensure that those payments move reliably, cheaply, and fast across a fragmented multi-chain environment. That is a stablecoin infrastructure problem.

On fast-finality networks, transfers can reach irreversible finality in sub-second time, reducing the operational uncertainty that comes with probabilistic settlement models. Stablecoin rails also support continuous settlement, including weekends and holidays, which reduces reliance on large pre-funding buffers in correspondent and multi-hop settlement. Research on stablecoins in agentic commerce notes that stablecoin-native batching and gasless execution make sub-cent payments realistic — but only when the liquidity layer supports them without friction.

But most stablecoin liquidity is still siloed across dozens of separate chains. An agent running on Arbitrum needs to pay a service operating on Base. A workflow on Solana needs to settle against a data provider on Polygon. Without efficient cross-chain routing, every one of these interactions requires manual bridging or a centralized intermediary, introducing exactly the kind of friction that agentic commerce is trying to eliminate.

This is the infrastructure gap that intent-based routing protocols are built to close. Eco's programmable stablecoin execution layer encodes money movement that reacts to conditions in real time, routing stablecoins across chains, vaults, and DeFi protocols with atomic settlement guarantees. Unlike traditional bridges that pass messages without execution guarantees, Eco's intent-based architecture ensures that a cross-chain transfer either completes exactly as specified or reverts entirely — no partial states, no bridge limbo.

Eco Routes v2 extends this further by supporting non-EVM chains through universal encoding. This matters directly for MPP-compatible payments: as services in the payments directory operate across Solana, EVM rollups, and other environments, a routing layer that handles chain-agnostic execution becomes necessary substrate. Routes v2 separates execution and reward deadlines, meaning solver economics are not held hostage by user urgency — a design that supports the low-latency requirements of agent session payments.

There is a second infrastructure challenge that session-based payments like MPP surface: liquidity provisioning. When an agent opens a session and commits funds upfront, those funds need to be available in the right asset on the right chain. If liquidity is fragmented or subject to slippage, the economics of micropayments stop working.

Shallow liquidity at the destination chain means higher effective costs for every session. That compounds across thousands of concurrent agents, making the whole system more expensive than the use case can sustain.

Eco's stablecoin liquidity layer, Crowd Liquidity, addresses this directly. It lets chains, bridges, and DeFi protocols share liquidity through a programmable, permissionless pool, reducing the capital fragmentation that makes cross-chain micropayments expensive. For developers building on top of MPP who need agent sessions to settle efficiently regardless of destination chain, the shared liquidity infrastructure of this kind removes a significant scaling constraint.

Crowd Liquidity is also designed to be collateral-free for solvers and permissionless to deploy, enabling it to scale to support a large number of concurrent agent sessions without requiring each solver to maintain a separate pool of capital per chain.

If you are building an application that involves AI agents performing tasks that require payment, MPP gives you a protocol-level way to handle those payments without building custom billing flows. Integrating an MPP-compatible wallet and implementing the session primitive means your agent can access any service in the payments directory without additional setup per service.

The practical implications extend across several categories:

Pay-per-call APIs can monetize without subscription tiers. Developers pay only for what agents consume. Services listed in the MPP directory — compute providers, data APIs, model inference endpoints — can receive micropayments without processing fees making small payments uneconomical.

Multi-agent workflows can allocate costs between sub-agents without centralized billing infrastructure. An orchestrator agent can open sessions for each service its sub-agents consume, settling continuously as work completes.

Autonomous treasury operations become practical when agents have defined spending limits and programmable controls. Programmable treasury functions, where workflows are delegated to agents with embedded controls like daily limits, transaction logic, and real-time auditability, are already in use at forward-looking organizations.

For developers who want to integrate cross-chain stablecoin payment flows without building routing infrastructure from scratch, the Eco Routes SDK provides a permissionless integration path that handles cross-chain execution, liquidity sourcing, and settlement. Agent wallets can settle MPP sessions through Eco-powered routes regardless of which chain the destination service operates on.

The intent-based architecture that underpins Eco is well-suited to agentic payment flows because it gives developers control over execution conditions without requiring them to manage routing and settlement logic themselves. Agents define desired outcomes — the solver network handles execution.

MPP is new, and the broader agentic payments stack is still in its early stages. Adoption of x402 has been limited so far, with most activity still in testing rather than production commerce. The same caution applies to MPP. Standards are only useful once services implement them and agents actually use them.

What is different in 2026 compared to a year ago is the specificity of the infrastructure being built beneath these protocols. Payment standards that had no viable settlement layer before now have purpose-built blockchain infrastructure, programmable liquidity networks, and developer tooling that reduces integration friction substantially.

Industries with high transaction volumes and thin margins — online advertising, cloud computing, machine-to-machine services — appear most likely to adopt token-based settlement first. Their reliance on automation and microtransactions aligns well with agent-driven payment patterns.

The combination of a well-designed protocol layer, a discovery directory of participating services, and cross-chain stablecoin infrastructure that can route and settle payments regardless of destination chain creates a more complete picture than any of these components alone. MPP provides the coordination language. Infrastructure like Eco's Routes and Accounts stack provides the execution layer. The question now is how quickly developers build services and agents that put both to work.

What is the Machine Payments Protocol (MPP)? MPP is an open standard, co-authored by Stripe and Tempo, that defines how AI agents and automated software can request, authorize, and settle payments for services without requiring human approval at each step. It supports multiple payment rails, including stablecoins and cards, and includes a session primitive for continuous streaming payments.

How is MPP different from x402? x402 embeds micropayment requests directly into HTTP using the 402 status code, making it lightweight and easy to implement for simple pay-per-request scenarios. MPP introduces a session model that allows continuous payment streaming with funds committed upfront, which is better suited to workflows involving sustained agent interactions with a service over time.

What blockchain does MPP run on? MPP currently runs on the Tempo blockchain, but the protocol is designed to be rail-agnostic. Visa has already extended it for card payments, and the specification is open for other blockchains and payment networks to implement.

Why do AI agents need stablecoins to pay for services? Traditional payment methods like credit cards are optimized for human-initiated checkout flows and carry minimum transaction costs that make micropayments uneconomical. Stablecoins settle in near-real time, operate continuously, and can be programmed with spending conditions, making them more compatible with the high-frequency, low-value transaction patterns that agentic workflows generate.

What infrastructure does MPP need to work cross-chain? MPP defines the coordination protocol but does not handle cross-chain routing itself. For agents operating across multiple blockchains, a cross-chain stablecoin routing layer that can move funds between chains with atomic settlement guarantees is needed to ensure that payment sessions can complete regardless of where the destination service operates.

What is the payments directory in MPP? The payments directory is a catalog of MPP-compatible services that agents can discover and transact with automatically. At launch it included more than 100 services spanning model providers, developer infrastructure, compute platforms, and data APIs. Service providers can list themselves in the directory to make their offerings accessible to any agent using MPP.