AI assistants are good at reading reports and spotting what changed. Giving them access to live ad accounts takes some care. Here is a safe way to do it.
What MCP is
The Model Context Protocol (MCP) is an open standard that lets AI assistants such as Claude and ChatGPT use tools: read a report, list campaigns, update a record. An MCP server offers the tools; the assistant decides when to call them.
For marketing, that means you can ask “which campaigns lost money last month and why?” and get an answer built from your actual accounts, not a guess.
The risks
- Too much access. Pasting an API key into an AI tool hands it everything that key can do, for as long as the key exists.
- Changes by accident. An assistant that can raise a budget or publish a tag can do it because it misunderstood a question.
- No record. If the assistant changes something, you need to know who asked, when and what.
Five rules for safe access
- Act as a person. The assistant should use the signed-in person’s own permissions, and lose them when the person does.
- Read-only by default. Allow changes only on the accounts where you want them.
- Preview before changing. Risky actions, such as spending more, publishing or deleting, should return a preview and wait for confirmation.
- Start paused. New campaigns and ads created by an assistant should start paused.
- Log every change, with the person and the time.
In Metric Sensei
The API/MCP Gateway gives AI tools one MCP server for every connected account, and follows all five rules. You sign in with your Metric Sensei account rather than pasting keys. The built-in assistant works the same way, and shows an approval card before anything changes.