Over the past few weeks, one acronym has appeared in virtually every adtech announcement: MCP. Amazon Ads launched its MCP server, Criteo followed with its first agentic campaign orchestrated with dentsu, while Adform, Equativ, StackAdapt and Meta have all stepped up their own AI connector initiatives. TikTok joined the list just last week.
“The AdCP era is over. Today, the market only talks about MCP servers.” Listening to some industry players, including the media buyer behind this comment, the debate would appear to be settled. The market has supposedly abandoned the agent-to-agent model that many proofs of concept focused on earlier this year in favor of a more pragmatic approach: connecting AI agents to existing tools such as DSPs and SSPs through MCP servers.
It is a punchy line, but also an oversimplification. More than anything, it illustrates the confusion still surrounding these technologies. MCP, AdCP and AAMP, the protocol recently launched by the IAB Tech Lab, do not address the same problem. Nor are they truly competing with one another. They simply operate at different layers of the future advertising architectures powered by AI agents.
A quick refresher for anyone who has not been following closely.
AdCP, or Agentic Campaign Protocol, is designed to allow agents representing buyers and sellers to automatically exchange information to negotiate, book or optimize media campaigns.
The protocol has already been used by Values.media to purchase a TV campaign, and by Olyzon and Swivel for CTV activation in the United States. Under this model, part of the commercial relationship is handled directly by agents.
“But to trigger actions or retrieve data, these agents need a standardized technical layer capable of connecting to the platforms across the ecosystem,” explains Raphael Ambit, founder of Epicflare, which supports adtech companies with their AI projects.
That is precisely the role of MCP. The Model Context Protocol acts as an infrastructure layer connecting AI agents to existing tools, data and functionality, including DSPs, SSPs, ad servers, CDPs, DMPs, data warehouses and analytics platforms.
“MCP is an effective way to make the programmatic infrastructure built before AI agent-ready.”
“MCP is like a universal plug that connects existing tools without fundamentally redefining their underlying business logic,” says Philippe Giendaj, Adtech Director at 366.
“It is a way to agentify—or make agent-friendly—traditional systems that were not originally built around AI,” adds Ambit.
The appeal is obvious: in many cases, connecting a platform becomes almost plug-and-play. “An MCP server abstracts away some of the technical complexity of the underlying APIs and allows an AI agent to interact quickly with legacy platforms without requiring a major integration,” explains Nicolas Cosson, co-founder of Concord.
This may be where the real shift is taking place. The industry’s main challenge was never getting agents to communicate with one another. “Multi-agent architectures have existed in AI for a long time,” notes Ambit. The central issue is making legacy advertising platforms usable within these new environments. He also points out that “AdCP was designed with the idea that this type of protocol would connect to MCP servers.”
As outlined above, the use cases are highly practical: automatically retrieving campaign data, reporting, performance audits, troubleshooting, budget optimization and campaign launches. All of this can be handled through prompts inside AI assistants such as Claude or ChatGPT, or through workflows orchestrated across multiple platforms.
“AdCP is trying to create a new agentic transaction channel, whereas MCP standardizes access to channels that already exist,” says Cosson. Those channels can be advertising-related, but not exclusively so, since every digital platform used by an advertiser—including AI, CRM, cloud and analytics systems—can develop its own MCP server.
This pragmatism explains the current focus on MCP.
For now, the market is primarily looking to unlock use cases that can be deployed quickly: automated reporting, data retrieval, troubleshooting, campaign activation and the orchestration of simple workflows.
Benoît Hucafol, SVP of Product Management at Equativ, makes the same point when he explains that the platform prioritized “the development of an MCP server from the outset, particularly because its first use case, curation, was not covered by AdCP.”
For an adtech company, opening an MCP server is therefore a relatively direct way to make its platform compatible with the AI assistants that traders, agencies and advertisers may use in the future.
By contrast, protocols such as AdCP or AAMP require much greater coordination among buyers, sellers, platforms and trusted third parties before they can achieve market-wide adoption.
“The market is not moving from one to the other. It is simply realizing that MCP solves an immediate and tangible problem, while AdCP addresses a more speculative and complex challenge,” Cosson continues.
The industry remains cautious about fully delegating commercial, regulatory or brand-safety decisions to AI. “If a buyer agent automatically negotiates with a seller agent and they each walk away with a different version of the outcome, which one is correct? Who is responsible for the final decision? Who settles a dispute?” asks Cosson.
Without a trusted third party or a robust reconciliation mechanism, it is difficult to imagine significant budgets being entrusted to fully autonomous protocols. These approaches can already support inventory discovery, prequalification and the automation of certain direct-deal processes. But fully automating agent-to-agent transactions remains extremely ambitious in the short to medium term.
That does not make AdCP or AAMP irrelevant—quite the opposite. These protocols aim to standardize media-buying operations. “AdCP, for example, defines standardized methods for inventory discovery, campaign booking and the structuring of advertising transactions,” Ambit explains.
Agent-to-agent protocols could eventually strengthen the sell side—and offline media
Their scope extends well beyond digital advertising. “AdCP redefines inventory booking through standardized methods and therefore does not apply exclusively to digital advertising,” Giendaj points out.
Its potential reach could extend to television, print, audio and out-of-home advertising. “It is notably a way to automate and standardize direct deals, which remain extremely time-consuming and poorly industrialized today,” Cosson adds.
The other issue is more strategic: the balance of power between the buy side and the sell side. Over the past several years, part of the value generated by programmatic advertising has shifted toward buying platforms and optimization layers. Agentic protocols could allow media owners and supply-side platforms to regain some control.
“AdCP also strengthens the supply side by giving sellers greater control over what they sell, to whom and at what price,” Giendaj argues.
Behind the technical debate surrounding MCP, AdCP and AAMP, the real question is therefore less about which protocol will win than about who will control the interfaces through which advertising inventory is accessed in the future.
MCP addresses the immediate need: connecting AI agents to existing tools. AdCP and AAMP are looking further ahead by seeking to structure advertising transactions in a world where an increasing share of operations will be delegated to agents. The market will probably need both.
