AI Agents Can Think. Now They Can Pay.
The internet's underlying infrastructure was built under the assumption that it would be used primarily by human users, but automated traffic now constitutes the majority of web requests. According to Cloudflare, automated systems account for 57.5% of HTTP requests as AI tools rapidly evolve into autonomous agents. Despite this transition, standard online payment interfaces remain tailored to human workflows, creating substantial friction for software. The Machine Payments Protocol (MPP) addresses this gap by offering a standardized payment interface designed for AI agents to settle transactions without human involvement. Machine Payments Protocol (MPP) is a standardized system co-authored by Stripe and Tempo to automate machine-to-machine payment negotiations without human intervention. Submitted to the IETF as the Payment HTTP Authentication Schema, MPP structures payment negotiation into three distinct steps: Challenge, Credential, and Receipt. These interactions occur via standardized HTTP headers, enabling software clients and agents to autonomously discover terms, submit proof of payment, and obtain receipts. Payment methods are governed separately by individual rails, ensuring MPP remains an open, fee-free foundation for internet agent transactions. The Machine Payment Protocol (MPP) workflow enables autonomous AI agents to pay for API and service access on demand using standard HTTP interactions. When an agent requests a service, the server responds with an HTTP 402 Payment Required status and a challenge header detailing payment terms. The agent verifies the terms against preconfigured delegated spending limits, signs the payment authorization, and retries the request with a payment credential. Upon verification, the server fulfills the request and provides a receipt, avoiding user account creation and preventing unpaid side effects.
閱讀原文 ↗目錄
The Problem with Today’s Payment Flow
The internet's underlying infrastructure was built under the assumption that it would be used primarily by human users, but automated traffic now constitutes the majority of web requests. According to Cloudflare, automated systems account for 57.5% of HTTP requests as AI tools rapidly evolve into autonomous agents. Despite this transition, standard online payment interfaces remain tailored to human workflows, creating substantial friction for software. The Machine Payments Protocol (MPP) addresses this gap by offering a standardized payment interface designed for AI agents to settle transactions without human involvement.
- Automated systems currently generate approximately 57.5% of HTTP requests to web content according to Cloudflare.
- AI tools are shifting from reactive conversational chatbots to autonomous agents capable of planning, executing, and evaluating tasks.
- Existing web payment systems remain constrained by human-centric design assumptions.
- Machine Payments Protocol (MPP) provides an interface that allows AI agents to settle payments directly with online services regardless of payment method.
What is MPP?
Machine Payments Protocol (MPP) is a standardized system co-authored by Stripe and Tempo to automate machine-to-machine payment negotiations without human intervention. Submitted to the IETF as the Payment HTTP Authentication Schema, MPP structures payment negotiation into three distinct steps: Challenge, Credential, and Receipt. These interactions occur via standardized HTTP headers, enabling software clients and agents to autonomously discover terms, submit proof of payment, and obtain receipts. Payment methods are governed separately by individual rails, ensuring MPP remains an open, fee-free foundation for internet agent transactions.
- MPP was co-authored by Stripe and Tempo and launched on March 18, 2026.
- The protocol core is published as the Payment HTTP Authentication Schema and submitted to the IETF standards track.
- MPP structures transactions around three core objects: Challenge (server request), Credential (client proof), and Receipt (confirmation).
- Protocol data is passed via HTTP headers including WWW-Authenticate: Payment, Authorization: Payment, and Payment-Receipt.
- Payment rails such as card networks, blockchains, and processors maintain their own specifications to interface with the core protocol.
- MPP is open and free, requiring no licensing fees to implement.
How Does MPP Work?
The Machine Payment Protocol (MPP) workflow enables autonomous AI agents to pay for API and service access on demand using standard HTTP interactions. When an agent requests a service, the server responds with an HTTP 402 Payment Required status and a challenge header detailing payment terms. The agent verifies the terms against preconfigured delegated spending limits, signs the payment authorization, and retries the request with a payment credential. Upon verification, the server fulfills the request and provides a receipt, avoiding user account creation and preventing unpaid side effects.
- Servers issue an HTTP 402 response containing a WWW-Authenticate: Payment challenge with terms, amounts, and recipient details.
- Agents authorize payments using delegated signing keys that have predefined spending caps, expirations, and recipient scopes rather than relying on LLM reasoning.
- Requests are retried with an Authorization: Payment header containing a single-use bearer credential that servers verify before executing any stateful actions.
- Failed verifications return a new HTTP 402 status with structured problem details (e.g., payment-insufficient, payment-expired) rather than an HTTP 401.
- With payment rails like Stripe, payments map directly to primitives like PaymentIntent without requiring customer account creation or API key purchases.
Two use cases of agentic payment
This section serves as an introduction to practical applications of MPP in the context of agentic payments. It signals a shift toward examining concrete scenarios where automated agent payment mechanisms can be utilized. The brief introductory remark frames upcoming examples illustrating these implementations.
- The section focuses on identifying use cases for MPP.
- The broader topic addresses applications of agentic payment systems.
- Specific implementation scenarios for agentic payments are framed for subsequent examination.
Use Case 1: How AI Agents Can Pay Pennies
AI agents frequently require granular, low-cost services such as web searches or lookups, but standard financial systems make sub-cent transactions unviable due to flat fees and confirmation delays. Traditional internet billing bundles services into subscriptions or credit packs to overcome this threshold, which does not suit the ad-hoc patterns of autonomous agents. To solve this, MPP utilizes session-based payments where an agent deposits funds upfront and issues signed IOUs per request. Once the session terminates, the accumulated total is settled in a single on-chain or banking transaction, lowering transaction costs and verification latency.
- Moving micro-amounts through traditional banking cards or blockchains incurs flat fees that often exceed the transaction value itself.
- Traditional billing relies on subscriptions and bundled credits because single micro-transactions fall below an economically viable cost threshold.
- Autonomous agents execute granular, continuous tasks (such as thousands of searches or lookups) that cannot be efficiently bundled into conventional plans.
- MPP addresses this friction by introducing sessions where money is set aside as an initial deposit.
- Within an MPP session, every individual request is accompanied by a signed digital IOU rather than an immediate financial settlement.
- Accumulated IOUs are settled in a single aggregate transaction when the session closes, amortizing the processing fee across thousands of requests.
Use Case 2: Purchase ByteByteGo Newsletter Single Article
Micropayment protocols (MPP) enable frictionless pay-per-use transactions, allowing readers to purchase single articles from publications such as ByteByteGo without recurring subscriptions. Demonstrated by Michael Blau using Drip and a Tempo wallet, writers can receive payments as low as a single cent in milliseconds. Looking forward, MPP provides a native protocol-level payment layer for the web, presenting an alternative to ad-driven monetization. Consequently, AI agents can become direct economic participants capable of autonomously purchasing services like compute and data.
- MPP enables pay-per-use micropayments for individual articles, removing the necessity of long-term subscriptions.
- Michael Blau demonstrated single-article purchases for ByteByteGo via Drip with agents transacting through an attached Tempo wallet.
- Writers can receive micropayments as low as one cent within milliseconds behind the scenes.
- Protocol-level micropayments offer a native payment layer for the internet as an alternative to advertising-funded models.
- Autonomous AI agents can use these payment rails to buy compute, data, and other services directly.
When Every Customer is a Stranger
Adopting the Machine Payments Protocol (MPP) eliminates traditional user signup flows, which complicates customer identification, abuse mitigation, dispute resolution, and commercial sales tracking. Because an MPP transaction exposes only a public key rather than user identity, identity verification and dispute handling are being developed as separate protocol layers. To ensure security and control, the MPP specification introduces strict guardrails, including mandatory TLS, expiring proofs, capped delegated signing keys, payload verification, and advisory-only service directories.
- Under MPP, payments only verify control of a public key, removing traditional account records like names, emails, and purchase history.
- Removing user accounts creates challenges for abuse control, sales conversions from free tiers, and customer contact.
- Identity and dispute mechanisms are intentionally designed as separate layers outside of the base MPP payment flow.
- Identity specifications backed by Visa and Cloudflare allow agents to sign requests so servers can recognize automated clients across workflows.
- MPP does not define a standard refund flow, relying instead on returning unclaimed session funds or refunding directly to the originating payment key via card networks or blockchains.
- MPP mandates security guardrails such as mandatory TLS, single-use proofs, no credential logging, capped delegated keys, and strict verification of signed payment payloads over descriptive text.
Your Next Customer Might Be an Agent
The Machine Payments Protocol (MPP) has been operational in production since March 2026, logging roughly 30,000 transactions by August 2026. While current transaction numbers are comparatively modest, early adoption resembles the formative stage of the Apple App Store, where early volume underrepresented future platform impact. The key significance is the arrival of automated agents acting as an entirely new category of paying customer.
- The Machine Payments Protocol (MPP) entered production in March 2026.
- As of August 2026, approximately 30,000 transactions have been processed via MPP.
- Early MPP transaction volumes are analogized to the initial launch phase of the Apple App Store.
- The protocol facilitates commerce driven by automated agents as a new class of customer.