MoonPay announced on May 1, 2026 that it is launching MoonAgents Card, a virtual Mastercard debit card that allows both users and authorized AI agents to spend stablecoins directly from on-chain balances at any online merchant that accepts Mastercard.
The payment infrastructure for agentic AI has been an open problem for long enough that the question was starting to feel theoretical. AI agents can browse, write, reason, and execute workflows, but the moment they need to actually spend money on a user’s behalf, the rails have not been there. MoonPay is making a direct attempt to close that gap. Built in partnership with Exodus and Monavate, the MoonAgents Card gives agents a programmable spending instrument backed by on-chain stablecoin balances, accessible through MoonPay’s agent infrastructure, CLI, and MoonAgents workflows, rather than through the consumer card flow that was designed for humans sitting at a checkout screen.
The practical significance of that distinction is easy to understate. A conventional virtual card can technically be used by an agent if you hardcode the credentials into a workflow and hope the merchant’s checkout flow does not trigger fraud detection or a CAPTCHA. That is not a product. It is a workaround. What MoonPay is describing is a card architecture designed from the start for programmatic spending, where agent authorization, spending controls, and transaction management happen through developer tooling rather than through a cardholder interface. The Mastercard network acceptance is what makes it immediately useful: an agent spending through MoonAgents Card can transact anywhere online that accepts Mastercard, which is effectively the entire e-commerce ecosystem.
The choice to build on stablecoin balances rather than fiat bank accounts is not incidental. Stablecoins settle faster, move across borders without the friction of correspondent banking, and can be held and managed on-chain without requiring a traditional banking relationship as the underlying infrastructure. For AI agents operating across jurisdictions, executing workflows on behalf of users in different countries, or managing spending at a cadence that would be unusual for a human cardholder, stablecoin rails offer a flexibility that fiat-native cards cannot match.
There is also a composability argument. Stablecoins live in the same on-chain environment as the DeFi protocols, smart contracts, and token ecosystems where agentic AI is increasingly being deployed. An agent that earns yield from a DeFi protocol, receives payment for a completed task in USDC, or manages a treasury on behalf of a DAO can now spend from that same on-chain balance through a Mastercard-compatible instrument without converting to fiat and back again. That closed loop, from on-chain earning to on-chain spending to merchant acceptance, has not previously existed in a form that a non-technical user could authorize and a developer could build on top of.
MoonPay’s existing position in the crypto payments stack makes this a natural extension rather than a pivot. The company built its reputation as the most widely used fiat-to-crypto on-ramp, embedded across hundreds of wallets and exchanges. That distribution gave MoonPay relationships with both the developer community and the consumer-facing platforms where crypto users actually live. Exodus, the wallet partner on the MoonAgents Card, is one of the most recognized self-custody wallet brands in the market, which adds a consumer trust layer to what is fundamentally a developer-facing infrastructure product.
The emerging market for agentic commerce infrastructure
The use cases for agent spending are more immediate than most people currently appreciate. Software subscriptions that need to renew without human intervention. API credits that an agent purchases as it scales a workflow. Travel or logistics bookings made programmatically as part of a business process. Cloud computing resources provisioned on demand. Each of these represents a category where human checkout is a bottleneck, where the friction of logging in, entering payment details, and manually approving each transaction slows down workflows that are otherwise fully automated. MoonPay is positioning the MoonAgents Card as the payment layer that removes that bottleneck.
The spending control dimension is worth examining separately. Consumer cards rely on the cardholder’s own judgment as the primary fraud and misuse filter. Agent cards need a different model, one where the developer or user who authorized the agent can set programmatic limits on what the agent is allowed to spend, on which merchant categories, at what frequency, and up to what per-transaction ceiling. MoonPay’s CLI and MoonAgents workflow integration is designed to surface those controls at the infrastructure level, so spending governance is built into the agent’s operating environment rather than bolted on afterward.
Visa and Mastercard have both been watching the agentic payments space closely, and Mastercard’s willingness to be the network behind this product signals that the major rails are not planning to resist agentic commerce. They are planning to participate in it. That network-level openness removes one of the potential regulatory and infrastructure barriers that might otherwise slow adoption.
The competitive race to own this transaction layer is just beginning. MoonPay is first with a product that is explicitly designed for agent spending on stablecoin rails with Mastercard acceptance, but the combination of factors that makes this compelling, programmable authorization, on-chain funding, global merchant acceptance, and developer-native controls, is visible to every other company working at the intersection of crypto infrastructure and AI payments. The window for establishing a default position in agentic commerce infrastructure will not stay open indefinitely, and MoonPay has moved to claim it while that window is still wide.
Also read: Roblox is launching AI development tools that put it in direct competition with Unity and Unreal Engine • PFlash claims a 10x prefill speedup over llama.cpp and it points to where local AI inference is heading • Morgan Stanley’s biggest Caterpillar bear just doubled his price target because AI data centers need generators