Alchemy is extending its AgentCard payment and identity platform with Mastercard Agent Pay, giving developers a new way to let artificial-intelligence agents complete online purchases using payment credentials tied to a consumer’s existing Mastercard. Announced September 17, the integration is designed to address a central problem emerging as AI systems move beyond answering questions and recommending products: how software agents can actually pay for goods and services while remaining constrained by explicit user authorization.

Under the integration, AgentCard can provide an AI agent with one-time-use, tokenized Mastercard payment credentials rather than exposing a consumer’s underlying card details directly to the software agent. Alchemy said those credentials can be linked to an existing Mastercard account, allowing users to retain applicable rewards programs, credit lines and other card benefits rather than opening a separate financial account solely for an AI agent. The arrangement gives developers access to conventional card acceptance while Mastercard’s payment controls remain part of the transaction chain.

Alchemy is presenting AgentCard as more than a virtual-card product. The platform combines several functions that an autonomous agent may need to operate across online services: payment credentials, a dedicated email address, a phone number and a stablecoin wallet. Developers can provision the package through Alchemy’s command-line tools, with the company saying the process can be completed in under a minute. The objective is to reduce the amount of infrastructure developers need to assemble before an AI agent can sign up for services, authenticate itself and transact on behalf of a user.

The Mastercard component expands the payment options available through that stack. According to Alchemy, the one-time credentials are intended for online merchants that already accept Mastercard but may not yet support newer agent-specific payment protocols. That matters because the emerging agentic-commerce market is developing faster than merchant checkout infrastructure can be replaced. An AI agent may be capable of comparing products, making a decision and initiating a purchase, but it still needs a payment method compatible with the merchant’s existing systems.

AgentCard therefore operates as a bridge between established card networks and newer programmable payment rails. Its stablecoin wallet gives agents access to blockchain-based and agent-native transactions where those mechanisms are supported, while tokenized card credentials provide a path through conventional e-commerce checkout systems. Alchemy has described its broader agent infrastructure as supporting both card-based spending and payment protocols designed specifically for software agents.

The structure reflects a broader fintech challenge created by autonomous software. Traditional payment systems generally assume that a person or business directly initiates a transaction. Agentic commerce introduces another participant: software that receives instructions from a user, decides how to fulfill them and may execute multiple actions before returning a result. That creates new questions around identity, permission, authentication, liability and the evidence needed to demonstrate that a payment was within the scope of the user’s instructions.

Mastercard has been building Agent Pay around those issues. Its framework emphasizes agent credentialing, programmable permissions, transaction execution and settlement, with Verifiable Intent designed to establish that an agent’s action corresponded to authenticated user instructions. Mastercard has said its machine-payment architecture is intended to support card, account and stablecoin settlement while allowing organizations to set authorization rules and spending boundaries programmatically.

Alchemy said AgentCard is being designed to support that Verifiable Intent framework. In practical terms, the system is intended to give participants in a transaction evidence of what a user authorized, the conditions attached to that authorization and which agent executed the action. For example, an authorization could contain limits on the merchant, transaction amount or other purchase conditions. That additional context is increasingly important when the software executing a transaction has discretion over how to satisfy a broader request.

The companies are also emphasizing consumer and issuer controls. AgentCard allows restrictions such as purchase limits, merchant-category rules and geographic or transaction boundaries to be defined for an agent. Those controls are intended to constrain an agent before the payment reaches the final authorization stage rather than relying solely on users to review completed transactions afterward. Alchemy’s current AgentCard service also promotes real-time purchase tracking and configurable spending limits for users connecting an existing Mastercard.

That control layer addresses one of the biggest adoption questions facing autonomous purchasing technology. Consumers may already use generative AI to search for products, compare alternatives or plan trips, but handing software authority to spend money creates a materially different risk profile. The payment system must distinguish between a recommendation, an instruction to prepare a transaction and an explicit authorization to complete it. Mastercard has made consumer choice and permission a central part of its Agent Pay strategy as it develops infrastructure for agent-led commerce.

Potential AgentCard use cases include relatively familiar consumer tasks. Alchemy says developers could create agents capable of ordering food, booking travel, renewing subscriptions or making other online purchases within predetermined rules. An agent could, for example, receive a budget and travel requirements, evaluate available options and then complete an eligible booking without requiring the user to manually enter payment details at checkout. The payment credential would be generated for the authorized transaction rather than giving the agent unrestricted access to the consumer’s primary card information.

For businesses, the integration offers another route to add transaction capability to AI products without building an entire payments and identity stack internally. Alchemy says companies can either use AgentCard as a ready-made experience or integrate its capabilities more deeply into their own applications and workflows. Developers can also build agents on different AI model providers rather than tying the payments layer to a particular large-language-model platform.

The Mastercard announcement follows Alchemy’s earlier expansion of AgentCard through Visa Intelligent Commerce. In June, Alchemy introduced an AgentCard integration that could provision Visa payment tokens alongside identity credentials and a crypto wallet. At the time, the company described its architecture as a routing layer capable of using emerging agent-native payment mechanisms where available and falling back to conventional tokenized card payments when merchants did not support them.

The Mastercard integration broadens that approach and illustrates how Alchemy is positioning AgentCard as network-flexible infrastructure rather than a payment product tied to a single rail. For developers building autonomous applications, that distinction is significant. An agent designed to perform tasks across the open internet will encounter merchants, subscription services and marketplaces with widely different payment capabilities. A system capable of accessing more than one payment network while also supporting stablecoin transactions may allow the agent to choose a compatible method without developers redesigning the application for each merchant.

Mastercard is simultaneously expanding its own infrastructure around agentic commerce. The company launched Agent Pay for Machines in June with support from more than 30 technology, payment and digital-asset companies, including Alchemy, Adyen, Coinbase, Stripe, Checkout.com, Global Payments and others. Mastercard described the initiative as infrastructure for environments in which software agents and connected machines could execute transactions programmatically and potentially at much higher frequency than conventional human commerce.

More recently, Mastercard has also been extending merchant-facing tools intended to connect retailers with AI-driven shopping environments. Its Agent Suite for Merchants and Agent Connect initiatives are designed to help merchants expose products to AI systems, connect third-party agents and support authorized transactions while maintaining control over their commercial rules. Those merchant-side initiatives complement payment products such as Agent Pay by addressing what happens before an agent reaches checkout: product discovery, merchant-agent connectivity and the rules governing access to commerce systems.

The combination points toward an emerging payments architecture in which different layers handle different responsibilities. AI models determine what action to take; agent platforms provide software identity and execution tools; payment networks provide credentialing and transaction infrastructure; issuers enforce financial controls; and merchants decide which agents and interfaces can access their products. Fintech providers operating between those layers are attempting to make them interoperable without requiring users to understand the underlying complexity.

Tokenization is especially important in that model. Giving autonomous software direct access to reusable card numbers would increase the consequences of software errors, compromised agents or poorly specified instructions. Single-use or limited-use credentials can narrow the permissions granted to an agent and reduce the value of a credential outside its intended transaction. Mastercard has framed network tokens and agent credentialing as core elements of its trust model for agentic commerce.

Verifiable authorization is another emerging requirement. Conventional card systems can authenticate the cardholder, but autonomous commerce requires systems to capture additional information about the relationship between the user and the agent. If a consumer tells an agent to purchase a particular product below a specified price, the payments infrastructure may need to preserve evidence that the final transaction matched those parameters. Mastercard’s Verifiable Intent concept is aimed at creating that additional layer of transaction context.

AgentCard also highlights the convergence of traditional fintech infrastructure and digital assets. Alchemy developed its business around blockchain infrastructure and continues to provide wallets and onchain developer tools, but AgentCard combines that technology with mainstream payment-network credentials. Rather than assuming autonomous agents will use only cryptocurrencies or only cards, the product is structured around access to multiple payment mechanisms. That approach reflects the fragmented state of agent payments, where card networks, stablecoins and newly developed machine-payment protocols are evolving in parallel.

The commercial opportunity will depend on more than technical payment compatibility. Merchants and issuers will need reliable methods for recognizing authorized agents, managing fraud and disputes, handling returns and determining when a user rather than the software should be asked to reconfirm a purchase. Consumer adoption will also depend on whether the convenience of delegating a transaction outweighs concerns about granting software financial authority. The controls embedded in AgentCard and Mastercard Agent Pay are designed to address those issues, although adoption will ultimately depend on how those protections operate across real-world transactions.

Alchemy’s September 17 announcement nevertheless marks another concrete step toward moving autonomous agents from product discovery into transaction execution. AgentCard is currently being offered as a live product, and its Mastercard integration gives developers access to existing online card acceptance while retaining a path to stablecoins and emerging agent-specific payment standards. According to Alchemy, the objective is to let developers give agents the infrastructure required not only to reason about a purchase, but to authenticate, pay and operate within explicit user-defined limits.

For the fintech sector, the more consequential development is the formation of a new payments layer around software actors. Agentic commerce requires payment credentials that can be delegated safely, permissions that machines can enforce, evidence of user intent and settlement infrastructure that works across existing merchant systems. By integrating Mastercard Agent Pay into AgentCard, Alchemy is attempting to package those requirements into a developer-facing product while relying on the reach of established payment rails. Whether autonomous purchasing becomes a mainstream consumer behavior remains uncertain, but the infrastructure required to support it is increasingly moving from prototypes into commercially available payment products.