Tools: reads & writes

Read tools

Reads return live context and have no side effects. They’re unrestricted — your agent can call them freely to diagnose. For broad questions, prefer runScript; the typed point-query tools (list campaigns, get insights, etc.) are for narrow lookups.

Write tools

Writes propose a single, scoped change — pause, enable, budget, bid, rename, create — and return a canonical envelope:

{ success, action, entityId, before, after }

The before/after snapshot is your confirmation that the change landed, with no follow-up read needed. Your client shows the proposal first; nothing executes until you approve.