Skip to main content

Activity and audit log

Everything an agent does through the MCP server is recorded — allowed calls and refusals alike — and surfaced in Core Web under Settings → AI connectors → Activity.

What each entry records​

  • which tool was called, with its parameters,
  • the outcome: executed (with the transaction id) or refused (with the policy rule that blocked it),
  • timestamp and the account involved.

Reading it as a safety instrument​

The refusals are as informative as the successes:

  • Refused calls you didn't expect — an agent repeatedly bumping the spend limit or probing networks outside the policy is misbehaving or mis-prompted; tighten the policy or disconnect it.
  • Successes you don't recognise — treat like any unauthorised transaction: disable the policy first (instant), then rotate the connector's access from your assistant and review revocation.

Transaction ids in the log link to the relevant explorer, so agent activity is verifiable on-chain like any other wallet activity.