Give your agents the ad accounts, without giving them the keys.
Plug Twine into the agent stack you already run and let your own agents, not just ours, work the ad accounts: reading context, preparing campaigns, staging changes. Everything they do is previewed, approved, and on the record, so nothing breaks quietly. In controlled beta today.
One command, and it’s part of the stack you already run.
No new platform to adopt and no workflow to migrate. Twine ships as a CLI and a local MCP server, so it registers into Claude Code or Codex the same way the rest of your tools do, and the ad accounts show up as one more thing your agents can work with.
From there, whatever you’ve built, a reporting agent, a Slack digest, a campaign copilot in your editor, can pull clean campaign context straight from Twine instead of CSV exports and dashboard scraping.
Your agents manage the campaigns. Ours are optional.
You do not have to run paid media through Twine’s agents to get the governance. Point the agent you already built at the account: it can read the state of every campaign, turn a brief into staged work, and make bounded edits like pausing a campaign, all through the same reviewed path.
Hand it the brief and go do something else. The head of growth gets work to approve instead of work to do, and nothing your agent stages can ship itself.
An agent on the raw ads API breaks things without telling you.
The platform APIs have no preview, no approval, and no undo. An agent working them directly will eventually change a bid strategy, a budget, or a status you did not intend, and you find out from the results. Through Twine, the same agent gets its work previewed before it lands, approved when it matters, attributed to its key, and reversible from a snapshot.
And because Twine watches the accounts themselves, anything that does touch them from outside is detected, attributed, and undoable. Nothing breaks quietly, including the things you did not build.
The agent accelerates the work. Twine still governs the decision.
This is the part that gets your agent project a yes. Access is a scoped, expiring, revocable key, not a login. Agents never hold ad-platform credentials, never approve or apply, and never get a second, less-governed path to the accounts. When security asks how the agent touches spend, the answer is: it doesn’t.
And because the CLI and MCP server call the same workflows as the app, everything your agents do lands on the same record as everyone else.
Get all of the leverage of agents on ad spend, none of the silent disasters.