Within 24 hours of each other this week, two ad tech platforms shipped the same kind of update: not a new dashboard or a new model, but a new door. Innovid opened its AI layer to Meta’s ads data through a shared protocol. Cint opened its research platform the same way, with a customer already running production workflows through it. Neither company built the other’s integration. Both used the same standard.
The protocol two platforms just adopted
The standard is the Model Context Protocol, or MCP, an open specification Anthropic published in November 2024 for connecting AI agents to external tools and data without a custom integration for every pairing. Until recently it lived mostly in developer tooling. This week it showed up twice in ad tech within a single news cycle.
On September 1, Innovid announced that NIVO, its AI orchestration layer, now connects to Meta’s ads MCP server, giving marketers a conversational way to pull Meta campaign data: summarizing campaign health, flagging performance trends, surfacing optimization opportunities and identifying creative that needs attention, all through natural language instead of switching between reporting dashboards. Innovid said it worked with Meta’s product team throughout the MCP server’s beta program.
“The promise of agentic AI isn’t just generating recommendations, it’s also enabling meaningful action across advertising workflows,” said Grant Parker, President at Innovid. “Our early work with ads MCP server demonstrates how secure, authenticated access to advertising data allows AI agents to analyze performance, understand campaign context, surface opportunities, and help marketers move from insight to execution.”
The same day, research and measurement platform Cint disclosed a parallel move: a Model Context Protocol server built with direct input from a customer, sample-and-panel firm Potloc, which is already running it in production. “AI is quickly changing how businesses access information, make decisions, and execute workflows, and we believe research and measurement need to be available within those environments,” said Lindsay Fordham, SVP Product at Cint. Potloc’s Director of Sampling Strategy, Mathieu Dussart, said the company’s fieldwork managers now “draft, price and validate target groups conversationally, in a fraction of the time.” Potloc reported that a single automated run set up a 20-market consumer banking study, complete with per-market locale, feasibility check and bid, in one pass, work that previously required configuring each market by hand.
Why one protocol beats a dozen point integrations
The significance here is not that two vendors added AI chat features. Ad tech has spent two years bolting conversational layers onto existing products. The significance is what those two announcements have in common: neither company built a bespoke, proprietary bridge to talk to an AI agent. Both plugged into the same open specification, which means a marketer’s AI agent of choice, in principle, doesn’t need a separate connector for Meta’s ad data versus Cint’s research data versus whatever platform ships an MCP server next.
That matters because Meta’s own ads MCP server only went into public release on July 16, six weeks before Innovid’s integration went live. A platform owner opening a standardized door, rather than requiring every AI vendor to negotiate its own custom access, is what turns a single vendor’s feature into an ecosystem pattern. Innovid’s own materials are notably specific about a limit worth flagging: the integration does not currently state whether NIVO can write back to Meta accounts through the protocol, only that it can read and surface campaign data. Read access and the ability to execute changes are different levels of trust, and vendors that blur the two in marketing copy are worth a second look.
Cint’s numbers carry a similar caveat. The 20-market setup claim is Cint and Potloc’s own reporting, with no third-party audit, no error rate, and no baseline comparison timeline disclosed. That doesn’t make the claim false. It does mean marketers should treat a vendor’s own efficiency statistic the way they’d treat any vendor benchmark: directionally useful, not yet independently verified.
What this means for the marketing leader
The practical shift is in how to evaluate the next AI agent pitch that lands in a marketing leader’s inbox. A vendor demoing a conversational assistant is not, by itself, differentiated; nearly every platform has one now, a trend this publication has tracked as AI copilots have converged across the major demand-side platforms. What is differentiated, and worth asking about directly, is the plumbing underneath: is the integration built on an open, durable protocol like MCP, or is it a point-to-point connection that breaks the moment either vendor changes its API. A protocol-based integration is more likely to survive a platform switch, extend to new data sources without a re-engineering project, and get maintained by more than one company’s roadmap.
Three questions are worth putting to any AI-agent vendor this quarter. First, is the integration read-only or does it have write access, and what approval sits between an agent’s recommendation and a live change to a campaign. Second, whose protocol is this built on, and does it survive if the vendor’s own AI strategy changes next year. Third, what independent evidence exists for any efficiency claim beyond the vendor’s own case study. None of that requires distrusting agentic AI. It requires treating “we added an AI agent” as the start of a procurement conversation, not the end of one.
Expect more of this. Once one major ad platform opens an MCP server, the incentive for every AI-layer vendor serving that platform’s advertisers is to plug in rather than build a proprietary workaround. The interesting story next quarter won’t be which company launched an AI agent. It will be which platforms opened a standard door for one, and which made vendors keep picking the lock.
Related: Every Major DSP Just Shipped the Same AI Copilot and Ad Tech Is Writing Two Rulebooks for AI Agents.
Source: Innovid
