LESSON 8 OF 8 · LAUNCH
Using AI assistants to run CTV campaigns: MCP, agents and controls
QUICK ANSWER
Some CTV platforms now let you plan, launch and change campaigns in plain language, through their own agent or through AI assistants connected over MCP. Set controls before you let an assistant touch budgets.
By The Tuning Desk · CTV Ad Finder · Published 2026-10-11 · Updated 2026-10-11 · 2 min read
What changes when an AI assistant can run campaigns?
Instead of clicking through a platform's screens, you describe what you want: a campaign for a region and audience, a budget shift, a report comparing two weeks. The assistant reads the platform's data and, if allowed, makes the change. For busy teams this saves time on routine work and makes reporting questions faster to answer.
What is MCP, and how is it different from a built-in agent?
The Model Context Protocol (MCP) is an open protocol that lets AI assistants connect to outside tools. A platform with an MCP server lets assistants such as Claude, ChatGPT or Gemini read its data and take actions the user approves. A built-in agent does a similar job inside the platform's own interface. The difference is where you work: in your assistant of choice, or inside the ad platform.
- Amazon DSP (now part of Amazon Ads Agent)
Amazon Ads Agent is described as a "conversational experience" to plan, create, optimize and measure campaigns in natural language.
Source: Amazon Ads: Amazon DSP
- Vibe (vibe.co)
Vibe MCP connects Claude, ChatGPT or Gemini to "Draft, launch, and adjust streaming TV campaigns" from the assistant, over OAuth, with edits logged. The MCP page says "Vibe MCP and the Vibe Developer Platform API are included with every Vibe account".
Source: Vibe MCP page
Which controls should you set first?
- Use the platform's own login and permissions, such as OAuth, rather than shared passwords.
- Start with read-only access for reporting before allowing changes.
- Require confirmation before any change to budgets, bids or targeting.
- Check that every change the assistant makes is logged with who approved it.
- Set spending limits in the platform that an assistant cannot raise.
Access is usually set up once by an account owner: you approve the connection in the platform, choose what the assistant may do, and can revoke it at any time. Check the platform's documentation for exactly which reads and writes are available, because they differ, and test with a small campaign or a copy before letting an assistant work on live budgets.
Where do assistants help most in CTV?
Pulling reports across campaigns, drafting campaign setups for review, checking pacing, and setting up test structures such as holdout or geo tests in a consistent way. Final decisions on budget, creative and measurement design should stay with people who own the results.
What are the risks?
An assistant can make a mistake confidently: misread a request, apply a change to the wrong campaign, or summarise a report incorrectly. In advertising those mistakes spend money. The controls above limit the damage, but people should still review what an assistant proposes before it runs, especially changes to budgets, bids, targeting and measurement settings. Keep a human owner for every live campaign, review the change log each week, and turn off write access for anyone who does not need it. Treat the assistant like a fast junior colleague: useful for drafting and checking, supervised for anything that spends.
What should you ask a platform?
- Which assistants can connect, and is access included or charged?
- Which data can the assistant read, and which actions can it take?
- How are actions logged and reversed?
- Does the platform hold a security attestation, such as SOC 2, covering this access?