Skip to content
2 minute readNewAgent-ready

Agents can work the approvals queue over MCP

Your own agent can now list, read, approve and reject held actions, and read the audit log, through the MCP server, the tool packages and the API. It can never decide a request it made itself.

The API reference expanded for the approvals endpoints, showing the approve operation and who may decide.

Held actions used to wait for a person at a dashboard. They can now be worked by the workspace owner's own agent as well, over the same API the dashboard uses.

  • Four new MCP tools. list_approvals (filter by status), get_approval, decide_approval (approve or reject, with an optional reason) and list_audit_events (who did what, including who decided each approval). The LangChain, AI SDK and Agents SDK packages get the same four.
  • Who may decide. A person, or the workspace owner's own key: an API key made without scope limits and not bound to an agent. An agent's own keys and tokens, and any key limited to scopes, can read the queue but still cannot decide.
  • Never your own request. The same agent or the same key that asked for an action gets a 403 if it tries to approve or reject it. Another owner key or a person has to.
  • Audit trail. Each decision records who made it (a person, or which key and agent) in the audit log.

Questions? Write to support@last-price.ai.