Markitech / Marketing MCP Server / Approval-Gated AI Actions
Approval-gated AI: nothing changes until you say so
Read freely, write only with approval. The rule that makes it safe to let an AI work inside live marketing accounts.
Updated November 4, 2026 · Markitech team
An AI that can change budgets, pause campaigns and publish tags is useful only if you can trust what it will do. The answer is not to stop it writing. It is to separate reading from writing, and to hold every write until a person reviews it.
In Markitech, reporting and lookups run automatically so you can ask as many questions as you like. Every write action, such as a budget change, a paused campaign or a Tag Manager publish, is proposed in chat and held. The hold is enforced by the system, so it still applies if someone asks the AI to skip confirmation.
How to do it
- Ask for the outcome you want, for example lowering spend on campaigns over target CPA.
- Review the proposed change list: which entities, which fields, old and new values.
- Ask follow-up questions or narrow the change before approving.
- Approve, and Markitech executes exactly what you reviewed.
- Check the result in the same conversation with a read-only follow-up.
Common mistakes
- Approving a change list without reading which accounts it touches.
- Bundling unrelated changes into one approval.
- Assuming a prompt instruction alone is enough protection on a live account.
- Forgetting to verify the change after it runs.
Ask Markitech
Related questions
Key terms
Approved by you, every time
Markitech proposes each change in chat and holds it until you approve. Nothing is published, created or changed in your accounts without that approval.
Part of the Marketing MCP Server guide.
Approval-Gated AI Actions questions
Can the AI skip approval if I tell it to?
No. The hold is enforced by Markitech’s system, not by the AI following instructions.
Do read-only questions need approval?
No. Reporting and lookups run automatically, so only changes to live accounts wait for approval.