Deploying MCP on Meta: a great promise, but how far should management be automated?

Managing your Meta campaigns in natural language: the promise of MCP

A new promise is stirring the market: deploying MCP to manage your campaigns. The Model Context Protocol is an open standard that connects AI assistants such as Claude or ChatGPT to tools and data sources. Applied to Meta’s Marketing API, MCP servers already make it possible to create, update and analyse Facebook and Instagram campaigns in natural language. Some of these implementations expose more than 70 tools covering the whole lifecycle of a campaign, from creation to automated optimisation rules.

The promise is strong and enticing. Tomorrow, part of the professions linked to media buying, management and consulting could be uberised, since everything would happen in natural language. At tigrz, we are investigating these tools. But we ask a simple question: how far will this automation go, and who does it really serve?

A promise that targets entire professions

The movement goes beyond a simple plug-in. Meta states the ambition of enabling companies to fully automate ad creation and targeting by the end of 2026. Mark Zuckerberg’s vision is explicit: a brand should be able to “just tell us the objective it is trying to achieve, how much it is willing to pay per result, connect its bank account, and we do the rest”. He speaks of a “redefinition of the advertising category”.

Let’s project ourselves 24 months ahead. I look at my campaigns, I ask Claude what the performance is, where to optimise, how to reprogram, which ideas to test. On paper, it’s all there. In practice, we think that will not be enough.
The real question: how far will slopification go?

We already have the practical example. The “wow” effect of the chat that writes our emails for us quickly wore off: many are coming back from it, for lack of flavour, relevance and impact. The risk is the same here. Does an automated analysis produce content without relief, a hodgepodge that is merely the weighted average of what should be said? Or do I have real advice in front of me?

That is the whole stake of slopification: by simplifying everything, you get a smooth deliverable, with no content, no position taken. Yet performance is not born from the average. It is born from the gap.

Why automation first serves the platforms

Things must be named. Entrusting management to full automation first comforts the platforms. The direction of history they tell is one of simplification: put everything in Advantage+, put everything in PMax, until any different programming becomes exotic. With fewer intermediaries, there will mechanically be less critical thinking, less capacity to question, to challenge, to take risks. We then settle for automated analyses.

But AI does not beat humans. AI is a tool, and you can see it well in campaigns. Analysing only through the lens of ROAS or CPA, these are KPIs: they do not say everything. They capture neither the courage of a singular vision, nor the instinct to step outside the frame to go after an impact on overall demand. And tracking biases do not make media optimisable by themselves.
What AI does not manage: context and courage

We today have the capacity to challenge and propose set-ups that are not ASC: full-funnel feedback, 360 integration, to generate search increments on Google, complementarity with other levers, even offline and ultra-local outdoor. These decisions are not for management done by Claude.

I agree: on first-level reporting, AI can work. And even then, it needs context. To understand a review, AI must be extremely well fed: what is the state of competition, how much stock is available, how does the strength of the commercial offer compare with last year’s. This whole context dimension drives optimisation and decision. Without it, reporting remains a number without reading.

Our conviction: intelligent insight produced by humans

Behind the promises of automation, there is therefore a real stake in serving the platforms’ interests. And we think the players who will perform best will be those who take risks, who innovate and who question performance in a 360 context. That is what the market will demand tomorrow.

Let’s be clear: it is not about refusing automation. From an agency’s standpoint, it is welcome, because it saves time and shifts effort towards strategy, advice and the client. But what advertisers, clients and CMOs will look for is intelligent insight produced by humans. Data-based, yes. But data-based does not mean data-automated. That is exactly where our value lies.

Keywords