Ads from an agent (MCP and API)
An agent connected over MCP can do with your ads what the Ads page does, with the same rules. Ads are off for every agent until you turn them on.
Turn it on
Open Developers. Each connected agent has an Ads switch: Off, See (read only) or Manage. API keys have the same switch. The consent screen tells you ads stay off until you pick one.
What it can do
| Read | Act |
|---|---|
ads_overview, list_ad_campaigns, get_ad_campaign | plan_ads (a plan, or a split test), nothing made |
ad_performance, ad_account_balance, get_ad_caps | make_ads (uses credits), send_ads_to_platform (lands paused) |
list_ad_suggestions, ad_tests, ad_alerts | boost_post (paused, no credits) |
ad_autopilot, ad_conversions, ad_audiences | pause_ad_campaign, set_ad_budget, decide_ad_suggestion, go_live_ad_campaign |
Money still waits for you
Turning a campaign on, raising a live budget, or approving an idea that spends more is always two calls. The first answers with a summary (the cost a day, the most this month, your caps) and a code. The agent shows you the summary. Only after you say yes does it send the code back, and the code works for ten minutes for that exact change. Lowering and pausing apply at once. No agent can raise a cap or turn on autopilot.
Every call an agent makes is in Activity with the agent's name, and in the Developers feed.
The same routes are on the API under /api/v1/programmatic/ads with an sk_live_ key: docs.txt, OpenAPI. The agent manual is SKILL.md.