How are AI assistants and APIs used to buy CTV ads?

From dashboards to conversations

For years, buying CTV meant working in a platform's web dashboard or asking an account team. Two newer routes are now published by several platforms: programmatic access through an API, and conversational access through an AI assistant.

What the Model Context Protocol (MCP) is

MCP is an open protocol that lets AI assistants connect to outside tools and data in a standard way. A platform that offers an MCP server lets an assistant such as Claude, ChatGPT or Gemini read data from it and, if permitted, take actions in it.

What platforms publish

On the pages we read:

• Vibe's MCP page says buyers can connect Claude, ChatGPT or Gemini to "Draft, launch, and adjust streaming TV campaigns". The assistant can read campaigns, strategies, creatives, audiences, metrics and inventory, and can create campaigns and edit targeting and budgets. Access uses OAuth and edits are logged. The page states that the MCP and the Developer Platform API are included with every Vibe account.
• Amazon describes Amazon Ads Agent as a "conversational experience" to plan, create, optimize and measure campaigns in natural language.
• Tatari describes an AI-powered Planning Engine, log-level exports and reporting APIs.
• MNTN lists "Flexible APIs (BI)" among its integrations and publishes AI bidding statistics.
• The Trade Desk describes its Kokai platform as analysing "up to 15 million ad opportunities each second".

What an assistant changes for a buyer

• Speed. Questions such as "which channels drove the lowest cost per visit last week" can be answered without building a report.
• Repeatable tests. A geo holdout or budget shift can be described in plain language and applied consistently.
• Lower barrier for teams without a trader. A marketer can make routine changes without learning every screen.

What to check before you connect one

1. Scope of access. Which data can the assistant read, and which actions can it take? Prefer read-only to start.
2. Authorisation. How access is granted and revoked, and whether it uses your existing login (OAuth) rather than shared keys.
3. Audit trail. Whether every change is logged with who or what made it.
4. Spend limits. Whether budgets and bids have guardrails the assistant cannot exceed.
5. Data handling. What leaves the platform when the assistant reads it, and under which terms.

How this shows up in our scores

Control and automation carries 15% of our weight. It rewards platforms whose pages describe how buyers can steer bidding, budgets and targeting, including through an API or an AI assistant. Data access through an API or exports is also one of the four Proof Card tiles.