Ad tech doesn’t suffer from a shortage of acronyms – but it does suffer from fragmented workflows.
Brand suitability provider Zefr is trying to fix some of that fragmentation by using Model Context Protocol (MCP) as the plumbing that connects AI agents to advertiser data and tools, together with an Advertising Context Protocol (AdCP) layer on top that allows those agents to turn natural language requests into live campaigns.
And now the jargon-free version: Advertisers can tell an AI agent what they want to run on YouTube in plain English, and the system figures out how to set up the campaign.
On Thursday, Zefr launched Zain, an agentic hub that serves as an entry point to that setup. The hub integrates data across platforms, including a brand’s first-party data, Zefr’s contextual data and data collected from third-party providers.
Through the AdCP integration, marketers can then push campaigns into social environments, including Meta and YouTube, without leaving the single workflow.
According to Zefr, the release marks the first time that YouTube campaigns are available to buy through an AdCP integration.
Nothing lost in translation
AdCP’s role in all of this is to translate natural language requests from marketers into the API calls that actually build and launch campaigns.
Basically, it creates a “common translation” that’s trained on a brand’s own terminology, said Jon Berke, Zefr’s SVP of platform development, so that advertisers can query an agent using their own internal terms and language, rather than preset verbiage.
Zefr customers have been able to test Zain since early April, but the latest version includes a new MCP server built into the hub that advertisers can access from within their own ecosystems, not just on Zefr’s platform.
The MCP server has two main elements, said Berke: a connection point to the cross-platform data and a set of instructions that translates natural language prompts and calls the right APIs.
The integration is meant to be straightforward to use, he said. Marketers just have to provide Zefr with basic account details and log-in info so the agent is able to access the correct data.
Go with the flow
Once that connection is in place, Zain handles the rest of the workflow.
Brands can launch a live YouTube campaign using AdCP by simply telling the agent that they want to run a YouTube campaign. The agent taps into Zefr’s AdCP, which contains instructions on how to build the campaign. If an advertiser neglects to share any important information, like budget or flight dates, the agent will follow up with the human before making the buy.
But the point of the agentic hub isn’t to “replace a buying platform” or disintermediate Google, which owns YouTube, said Berke. It’s in Zefr’s best interest to stay in Google’s “good graces,” he added.
The benefit is really for advertisers, he said, because they can execute campaigns in a single workflow “using the same tool that they use to execute in other parts of the ecosystem.”
If a Zefr client has its own local Claude instance, for example, they can just tell Claude, “I want to do a buy using Zefr data on YouTube,” without actually having to access Zefr’s platform, Berke explained. The entire process runs through the brand’s own agent, with Zefr’s historical data available as an additional training source beyond the brand’s data.
Where the magic happens
At launch, Zefr is using the integration mainly for Demand Gen campaigns, Berke said, which is Google’s top priority at the moment. Demand Gen is Google’s AI-powered campaign format that targets potential customers as they engage with their Google feeds, like checking email or watching YouTube shorts. Over time, he added, Zefr expects to support a wider range of standard video buys, including some that aren’t currently available through YouTube’s own API.
Typically, the only way to activate a YouTube buy programmatically is with a “really in depth understanding of how to build an automated activation,” according to Berke, which requires what he called some “back-end magic.”
Berke declined to explain what, exactly, that magic entails. It’s “our secret sauce,” he said.
What he did share is that the solution involves building translation layers, rather than new pipes. If a brand is working with an audience verification vendor, like Nielsen or Comscore, they can now combine that data with Zefr’s contextual insights to get more nuanced answers to questions like, for example, what sort of content indexes highly with viewers over 40.
In one case, an automotive brand client of Zefr’s running both English and Spanish campaigns noticed that the Spanish activations were getting far more engagement when the content was focused on food, Berke said. The brand developed new ad creative with a focus on drive-through meals and families heading to dinner together in their car. Zefr was able to bring in new, food-focused audiences as well.
That’s the kind of pattern Zefr wants to surface faster and plug directly into planning and activation.
Reducing fragmentation and bringing all of a brand’s insights into one place is “why we wanted to make [Zain] available as an MCP,” said Berke. “It’s very wide-ranging.”
