Mercury Launches Virtual Cards for AI Agents
Mercury launched Agent Cards on August 12, enabling AI agents to retrieve dedicated virtual-card credentials through its API or CLI and make online purchases within administrator-set controls. Mercury's support documentation states that agents cannot create new cards, increase their own spending limits, or access cards not explicitly assigned to them. The launch is part of Mercury Spend, a broader spend-management offering for existing business banking customers.
Mercury launched Agent Cards on August 12, a virtual-card type that businesses can assign to AI agents for autonomous online purchases. According to Mercury's product documentation, an agent can retrieve a card number, expiration date, and CVC through Mercury's API or command-line interface, allowing checkout without a person entering credentials.
The product is designed for tasks including vendor and procurement payments, flight bookings, online advertising, API usage, developer tools, and SaaS subscriptions. Mercury states that Agent Cards are accepted anywhere Mastercard is accepted online, and that both debit and credit virtual cards can be designated as agent cards.
Spending controls and credential access
Mercury's documentation distinguishes Agent Cards from ordinary virtual cards primarily through autonomous credential retrieval and restrictions on the agent's authority. Administrators set the spending limit and explicitly assign the card to an agent. The documentation states that an agent cannot create a new card, raise its own limit, or reach another card that was not assigned to it.
FinTech Global reports that Mercury Spend also lets administrators restrict spending by merchant or merchant category, track and audit activity, and cancel agent cards. It reports that transactions outside assigned parameters are declined. PYMNTS similarly reports that cards are created by a human, while the agent can use them autonomously within the relevant budget, merchant, and category controls.
These controls address a distinct operational issue in agentic systems: a model or workflow can be authorized to execute a task, but payment credentials turn that task into an external financial action. In comparable implementations, least-privilege credentials, scoped budgets, merchant restrictions, and auditable transaction logs are practical safeguards against prompt injection, flawed workflow logic, or overly broad tool permissions.
Part of Mercury Spend
The Agent Cards launch accompanies Mercury Spend, which FinTech Global describes as a spend-management product for issuing cards to employees and AI agents from the same banking dashboard. The publication reports that the offering also includes budget allocation, transaction categorization rules, and receipt collection through Gmail inbox scanning or employee text-message submission.
According to FinTech Global, cards can be automatically frozen when required receipts, memos, or transaction categories are missing. It also reports that spending information is displayed alongside cash position and runway. Mercury Spend is available to Mercury's existing business banking customers, the publication says.
Mercury CEO Immad Akhund told FinTech Global that "Companies can be smaller and faster than they were two years ago, and they're a mix of people and agents doing real work." In comments reported by PYMNTS from Fast Company, Akhund said customers had already been manually issuing Mercury virtual cards to agents, and characterized the new product as adding controls and auditable agent spending.
Implications for agent builders
For teams building purchasing agents, the announcement makes payment authorization a first-class integration concern rather than a credential-sharing workaround. Mercury does not build AI agents, according to its support documentation; customers can use their chosen AI platform, including Anthropic Claude or OpenAI products, while Mercury supplies the payment credential, controls, and oversight layer.
The practical boundary remains important: a payment card can constrain where and how much an agent spends, but it does not independently validate whether the agent's intended purchase is semantically correct. Companies deploying comparable systems commonly need application-level approval rules, vendor allowlists, idempotent transaction handling, and monitoring around the agent workflow in addition to card-level controls.
Key Points
- 1Mercury Agent Cards provide agents dedicated virtual-card credentials, enabling autonomous online checkout while separating their spending activity from human cardholders.
- 2Mercury documents hard card-level limits: agents cannot create cards, raise limits, or access credentials beyond explicitly assigned cards.
- 3Comparable agent-payment deployments require workflow-level safeguards because transaction controls constrain spending but do not verify purchase intent.
Scoring Rationale
The launch provides a concrete payments primitive for teams deploying agents that need to purchase services, tools, advertising, or travel. Its scoped credentials and audit controls are relevant to agent security design, although the product is limited to Mercury customers rather than a broadly available model or developer standard.
Sources
Primary source and supporting public references used for this report.
Practice with real Retail & eCommerce data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Retail & eCommerce problems
