By Magnus Ohlin, co-founder and Chief Innovation Officer at Camphouse
The Model Context Protocol (MCP) is fueling many fantasies about a finally unified advertising market. In reality, the stakes are elsewhere. It is less about simplifying than about systematising, by enabling data to be exploited at a level of granularity currently inaccessible.
Let’s be clear: a new attempt to unify the media market is not for tomorrow, and the MCP is already a demonstration of this. Several protocols are emerging in parallel, universality is questioned from the outset, and each player develops its own approach, environment, and logic. As Paul Ripart, Digital and Data Deputy Director at Prisma Media, points out, the risk is even to recreate walled gardens, but at a more invisible level.
Thinking that a cross-media campaign can be managed in a totally unified way is not realistic. The purchasing logics remain too different. The technical constraints, formats, and rules specific to each platform prevent any complete centralisation. This is not where the transformation lies.
The real potential: micro-tasks for micro-data
The true potential of the MCP is micro. It lies in the execution of micro-tasks, on a large scale: categorising actions, applying a coherent and enriched taxonomy, correctly generating UTMs, structuring campaigns according to naming conventions and predetermined fields, and placing the right information in each of the platforms.
These are simple tasks, but repetitive and massive. And today, they still largely rely on humans. The MCP allows them to be automated, made reliable, and executed uniformly. This is concretely the generation of structured data, micro-task by micro-task.
Unraveling to regain granularity
The problem today is that we have an extremely rich media fabric… but completely tangled. All the data is there. All the threads exist. But they are intertwined, superimposed, and difficult to read. To regain value, we must accept to unravel. Go back to the thread. Understand each campaign, each piece of data, each interaction. Regain granularity.
It is only at this level that data can truly be exploited. And this is precisely where the MCP becomes useful. Not to unify, but to activate these threads one by one, at the right scale. Once this work is done, we can reweave. Recreate a coherent, readable, and actionable network.
Moving from API logic to usage logic
Another key point is the ability to go beyond APIs. We focus a lot on MCP, and this is logical given its strong capacity to interconnect the media ecosystem. But agentic goes beyond MCP. The Claude Cowork and Claude for Chrome duo, which combines a Browsing Agent that takes control in your Chrome, and Claude Cowork which directly manipulates the APIs it has access to, is a total upheaval of the media ecosystem.
Today, many workflows rely on complex technical integrations, often limited, and each living in its own SaaS, each in its own browser tab. Agentic in its entire range, from MCP to virtual action-taking, allows for direct and universal interaction with all interfaces.
This enrichment of the agentic range makes us move from a technical logic to a usage logic. This complete agentic range completely modifies the way campaigns can be operated.
An essential discipline on data
But this mechanism only works if the data is clean. The MCP does not correct anything. It executes. Without strong data hygiene, it will simply reproduce and amplify errors. It is therefore necessary to impose a very clear discipline, which goes as far as very concrete rules: account presets, prohibitions, purchasing frameworks. A form of toll that imposes rigor in execution. Without this, automation is of no interest.
Concrete but targeted interoperability
The interest of the MCP is in interoperability, but not the one we imagine. It is not about creating a single system, but about allowing agents to coordinate categorisd actions between tools that live precisely in a plurality of platforms.
This can, for example, consist of connecting media planning with tools like Asana or Monday, automatically generating a media plan from a brief, assisting with the briefing, or distributing data within the platforms. Interoperability that allows working in the interface of your choice: creating a media plan from Claude within a media operations platform, or from this same platform by taking advantage of Claude’s agentic strength. The essential thing: respecting usages.
Better exploiting complexity, not eliminating it
The media market will not simplify. But it can be better exploited. The MCP and agentic AI more broadly with agentic browsing, make it possible to de-complexify operations, organise actions, optimise each purchase at the task level, and make data actionable.
Provided we accept a simple reality: we will not resolve complexity through unification. We will make it exploitable through structure. The subject is not to connect ever more, but to better organise what already exists. To avoid automating chaos.
