← Blog

How to Track and Audit AI Agent Spending Across Every Card

Issuing a separate card per task is the right way to contain what an AI agent can spend — but it creates a reporting problem. Once you have dozens of single-purpose cards, "what did my agents spend this week?" is no longer a question you can answer by staring at one card. Agentcard now gives you two account-wide views built for exactly this: a flat list of every transaction across all your cards, and a list of every app you have connected. Together they let you reconcile spend and audit access in seconds.

See every charge in one list

The list_all_transactions tool returns transactions across all of your cards in a single flat list, newest first. Each row is tagged with the card it belongs to — its id and last four digits — so you can always tell which card was charged. This is the account-wide companion to list_transactions, which still exists for when you want just one specific card.

From the CLI:

agent-cards transactions

Or ask your agent directly:

"List all my Agentcard transactions from this week and total them by card."

You can narrow the view with a few parameters:

  • status — filter to a state such as PENDING, SETTLED, DECLINED, REVERSED, EXPIRED, or REFUNDED. Pulling just the declines is a fast way to spot an agent hitting a card limit.
  • limit and offset — page through history (default 20, up to 100 at a time).

Because every transaction carries its card's last four, totalling spend per agent — when each agent owns a card — is straightforward. For the deeper story on automatic logging, see AI Agent Expense Tracking and AI Agent Expense Management.

Audit who can spend on your behalf

If you let third-party apps and agents connect to your Agentcard account over OAuth — a coding agent like Kilo, for example — you should be able to see exactly who has access. The list_connections tool lists every connected app, the OAuth scopes it was granted, when it first connected, and whether the connection is still active.

agent-cards connections
"Which apps are connected to my Agentcard account, and what can each one do?"

If something looks wrong — an app you no longer use, or a connection you do not recognize — revoke it immediately:

agent-cards connections revoke <clientId>

A simple audit routine

  1. Run list_all_transactions and skim the newest charges; pull the DECLINED set to catch agents bumping into limits.
  2. Total settled spend by card to attribute cost to each agent or task.
  3. Run list_connections to confirm only the apps you expect can spend, and revoke anything stale.

This pairs naturally with the least-privilege approach in Financial Zero Trust for AI Agents: limit what each card and each connection can do, then verify it with these two views.

Next steps