Skip to content
Automation Beta

Connecting MCP servers to AI: extra skills for your assistant

Yes. MCP stands for Model Context Protocol, an open agreement that lets a system offer its own skills to an AI assistant. Connect such a server to IntraGPT and the assistant can use those skills, while you decide which of them are switched on. The platform labels this part as beta and we set it up together with the customer.

Every organisation has a system its supplier says can do plenty, but that nobody can reach without logging in. More and more often that supplier offers an MCP server: a standard way for a system to describe what it can do and offer that to an AI assistant. Connect such a server and you no longer have to wait for a separate integration to be built first. In IntraGPT you register the server under a name and a description of your own. The skills it offers then come into view and you pick which ones the assistant may use. Pick nothing and everything is open, which is rarely wise with an extensive server. A server can also be switched off temporarily without losing its settings. The platform labels MCP servers as beta: we set the first server up together with the customer, look at which skills it contains and only open up what you are actually going to use.

What a connected MCP server delivers

Skills without custom work

If a system offers an MCP server, the assistant can use that system's skills as soon as the server is connected. Nothing has to be built to measure first.

On or off per skill

You decide which skills of that server the assistant may use. What you do not allow it cannot call, even if the server offers it. That is how you make a broad server as narrow as you want it.

In your environment or at the supplier

An MCP server can run inside your own environment or sit with the supplier of the system. You connect both in the same place, so the overview of what is switched on stays in one spot.

See what is inside first

Before anything opens up we map out which skills the server offers and check that the connection works. Only after that does an employee notice anything.

Switching on and off without fuss

Every connected server has its own switch. If a supplier does something unexpected or you want to pause for a moment, you switch that server off and the settings stay as they were.

Situations where an MCP server saves work

Industry and manufacturing

Situation
A supplier ships an MCP server with its planning software, and work preparation wants to be able to ask it questions.
What the agent does
The server is connected and only the skills that look something up are allowed.
Result
The assistant consults the planning without a separate integration having to be built first.

Government

Situation
A department has an internal tool that is connected to nothing else and that only a few people can operate.
What the agent does
The tool is connected to the assistant through its MCP server, with only the skills the department needs.
Result
What used to be a request to the administrator is now a question in the chat with an answer from that same tool.

IT operations

Situation
An MCP server offers dozens of skills, a few of which change things nobody wants touched by accident.
What the agent does
Those skills are switched off and the rest simply stays available.
Result
The assistant no longer sees them and cannot use them, while the lookup work carries on.

Customer service

Situation
A service desk wants to check a status during a call, in a system whose supplier already provides an MCP server.
What the agent does
The server is connected and limited to retrieving that status, nothing more.
Result
The agent asks the question in the chat and no longer needs the other system open alongside.

From server to usable skills

  1. 1

    Connect through the platform

    You register the server under a name of your own and connect it with a token that you issue yourself and can withdraw again.

  2. 2

    Set the permissions

    You choose which skills of that server are allowed and which roles inside the organisation may work with the assistant that uses them.

  3. 3

    What the assistant sees

    Only what you have allowed. The rest of what the server offers stays invisible to it, including anything the supplier adds later.

  4. 4

    Putting it to use together

    MCP servers sit in the platform as beta. We set the first server up together with the customer, look over your shoulder in the first days, and every action is recorded so you can see what happened.

Connected securely

What an external server may and may not do

An MCP server belongs to another system and offers skills to your assistant. The first question when connecting one is therefore not what it can do, but what it is allowed to do.

Where the data lives: Nederland of de EU

A token you withdraw yourself

The connection runs on a token you issue yourself. Want the connection to stop, then you withdraw that token and it is over. The token belongs to the server setting and never appears in a conversation.

Your choice sets the boundary

Only the skills you switched on are usable. If the supplier adds something later, it does not open up by itself and you get to decide about it first.

Every action recorded

What the assistant does through a connected server is recorded. Afterwards you can trace when something was retrieved and at whose request.

Own server, no training

The model itself runs on our own server in the Netherlands. What comes in through a connected server stays inside that setup and is never used to train a model.

Frequently asked questions about MCP servers

Does IntraGPT work with MCP servers?

Yes. You register an MCP server in the platform, choose which skills the assistant may use and switch it on. From that moment the assistant can work with it. The platform labels this part as beta and we set it up together with the customer.

What does beta mean here exactly?

It means the part sits in the platform as beta and that we set it up together with the customer. We connect the first server with you, look at which skills it contains and only open up what will actually be used. The way of working is the same as everywhere else in the platform, the guidance is more intensive.

Can I connect MCP servers to ChatGPT?

ChatGPT can handle MCP, but then the conversation and everything that comes back travels through a public service. With IntraGPT the administration stays in your own environment, you decide per server and per skill what is allowed, and the model runs on our own server. The same way of working, a different place where the data passes through.

Which MCP servers can I connect?

Any server that follows the protocol, whether it runs inside your own environment or sits with the supplier of the system. What that server offers decides what the assistant can do. During setup we map that out first, so you know what you are switching on before anyone works with it.

Can I exclude skills?

Yes, and it is the most important thing you configure here. You can make a short list of allowed skills, which closes off everything else, or switch off individual skills. Do neither and everything the server offers is allowed, which with an extensive server is rarely what you want.

What if the MCP server is slow or does not respond?

Then the conversation does not hang. A call that takes too long is cut off and the assistant reports that it got no answer, instead of waiting or inventing something. If a server stays unreliable you switch it off for a while without losing the settings.

What does connecting an MCP server cost?

That sits inside the project we run together. What drives the work is the number of servers, how many skills may be open and whether anything may be changed as well. We make that concrete in a thirty minute call.

How long does setup take?

Registering a server and switching it on is a quarter of an hour. The time goes into deciding which skills may be open and who may use the assistant. Because this part sits in the platform as beta, we deliberately plan a moment to look at the first usage together.

For which sectors

Other connectors

See all connectors

Curious what this connector would deliver for you?

Book a free thirty minute AI call. We look at your systems, the permissions around them and the first use case that saves time or money.

Book a free AI consultation