Will MCP kill applications?
MCPs are multiplying and AI generates dashboards on demand. Will applications disappear? We have no certainties, but a few hypotheses, one fear, artifact obesity, and some good news: AI and applications reinforce each other more than they compete.
Connecting intelligence, not data
What is an MCP?
Among seasoned AI users, MCP is already well established. For everyone else, here is a reminder. An MCP connects an AI to an application: the AI can access its data, sometimes act within it, and do with it whatever the user asks.
MCPs are multiplying. You can already plug your AI into your email, your drives, your bank accounts, or into the specific data an application calculates. And the AI turns it into whatever you want: an analysis, an agent, a dashboard…
We see it ourselves with the nostress MCP, open to our customers, who use it more and more. And this is only the beginning.
The awkward questions
So, inevitably, questions arise. Are applications doomed to be replaced by a multitude of AI dashboards? Will agents remove any point in humans using data? If an agent does my job, do I still need my application?
Let’s be honest: we are only at the beginning. Any answer today is a hypothesis, not a certainty. Here are ours.
Artifact obesity
The AI-generated dashboard is very often a super report, hyper-personalized. That is its strength, and also its limit: the more it is tailored to one person, the more it leaves everyone else out.
We have been here before. Reports multiplied until they created so much noise that the data could no longer be heard. With AI, we can produce ten times more of them, ten times faster. There is a real risk of heading toward artifact obesity.
nostress has no miracle cure for that. But there is one thing we can guarantee.
Source from KPIs, not from data
The nostress MCP is a KPI MCP. The indicators arrive already calculated, with a shared method. Everyone can build the dashboard they want, and everyone still sees the same information. And that is essential.
Plugging an AI straight into raw data means leaving everyone to calculate their own indicator. At best, it creates confusion: three people, three definitions of margin or service level, three different numbers in the meeting. And an indicator like service level cannot be improvised: it requires a real method, which everyone then risks reinventing in their own way. Not to mention that it costs a fortune in tokens, since the AI has to ingest and recalculate everything for every question.
This is even truer for predictive KPIs. A product’s future sales come out of a neural model trained for that purpose. Handing that calculation to a prompt, however powerful, is neither reliable, nor reproducible, nor economical.
So dashboards, yes: plug in, that is what it is for. But source from KPIs, not from data. That is our mission: connecting intelligence, not connecting data.
It is also why a shared application remains necessary. It guarantees a common view and brings everyone on board. In fact, we are seeing it: at nostress, the explosion of AI comes with a rise in application usage. That is already a first part of the answer.
What about agents?
Tomorrow, will information be meant for AI rather than for people? That is a harder question. Whatever people say, agentic AI is still very little deployed. And where it is, you don’t have to scratch far to see that its actual scope often remains very narrow compared with its potential.
But one thing is clear: an agent needs a lot of data, of good quality, and above all as many calculations as possible done upstream. An agent has better things to do than calculate a potential: it has to react to it. Like a person or their AI, it will quickly need intelligent KPIs more than data. We are still connecting intelligence, but this time from AI to AI, in service of a mission.
And if an agent does my job, do I still need my application? Let’s put the question another way: does a manager no longer need an application because they have delegated their work to a team of ten? Of course they do. Their need for access may even be greater. And that is what we are seeing.
The web did not kill the store
Today, you can see MCP as the application that will kill applications, just as people thought the internet would kill stores. The web grew very fast, then found its place. It did not kill stores, because it met a new need.
MCP may follow the same path. To what extent? We don’t know. But we are convinced it will never be 100%.
What is certain, however, is that the stakes in infrastructure, in the relevance of information and in keeping token costs under control are enormous. They change every day, and that is where we put our energy.
Everyone knows how to connect data. Connecting intelligence is something else.