Skip to main content

MCP server: connecting AI assistants

Your employees put their questions to an AI assistant, and the assistant fetches the answer out of projectfacts. Through the Model Context Protocol you connect Claude, ChatGPT or an agent of your own, without anyone being able to see more than before.

Enquire now

Can I connect Claude or ChatGPT to projectfacts?

Yes. projectfacts comes with an MCP server of its own. MCP stands for Model Context Protocol, an open standard through which AI applications reach external data and functions. Once the access is set up, an assistant can query data from projectfacts and, within the rights granted, create cases. Questions are asked in ordinary language, not in a query language.

The practical difference from a report: a report answers the question somebody anticipated when the report was built. An assistant answers the question you have right now. The protocol was presented by Anthropic at the end of 2024 and then published as an open standard, and the common AI applications have supported it ever since.

What the MCP server is good for

Questions instead of reports

Analyses come out of a conversation, even when nobody built a matching report beforehand.

Your rights still apply

An MCP access extends no permissions. Everyone sees exactly what they are allowed to see on screen.

Tailored access

Instead of one master access you set up several small ones, by department and by task.

The most important sentence first: there are no extra rights

With every AI connection, the first question is the one about the data. With MCP the answer is clear and easy to check: an MCP access can never do more than the user behind it.

What that means in practice

When you call a tool through the assistant, projectfacts checks your personal rights profile, exactly as it does when you log in on screen. Two employees using the same tool therefore get different results, each their own. So the fact that a tool is there says something about the range of functions, not about anything being released to you.

That takes much of the weight out of the introduction. You do not have to build a second permissions concept or maintain special rules. The rights system you have anyway continues to apply.

One access with all the tools compared with several tailored MCP access points | projectfacts

Several small access points instead of one big one

The MCP server of projectfacts is built in modules. You do not set up one access with all the tools, but several, each of them cut to one task.

Why that is the better design

Before every answer, an assistant has to work through the list of available tools. The longer that list, the more attention it takes up and the less precise the answer becomes. An access with twelve fitting tools delivers better results than one with a hundred, ninety of which have nothing to do with the question.

What such a set-up looks like in practice

Three to four access points are usual: a general read access for all employees, one with HR topics for personnel administration, one with customer and sales topics for CRM and, where it makes sense, one for project work. Which tools an access contains is set by an administrator and can be changed again at any time.

How the connection is set up

The route is short and needs no development work.

Three steps

First an administrator creates an MCP configuration in projectfacts and selects which tools it contains and which user groups may use it. After that, every employee enters the address of this access in their AI client. Logging in runs through OAuth, so the connection hangs on their own account and no passwords are handed on.

Which assistants can be connected

MCP is an open standard. Every application that supports it can be connected, among them Claude and ChatGPT as well as agents of your own that you run yourself. Because the standard is open, you connect projectfacts once and stay free in your choice of assistant.

What you should know about it

An MCP access belongs to one tenant. Anyone who works in several tenants sets up one access per tenant. If an administrator changes the tools of an access, the change takes effect as soon as the AI client connects again.

MCP or REST API?

Both routes lead out of projectfacts, but they solve different tasks. The rule of thumb: the REST API is for software that always does the same thing. MCP is for questions you do not know in advance.

How to tell which is which

If something is to run reliably and identically every night, the API is the right route. If somebody in a department should be able to ask a question that nobody knew about beforehand, it is MCP. The effort differs as well: MCP needs a configuration, while an API connection brings development, testing and operation with it.

In practice the two sit side by side: the API keeps the fixed processes running, MCP answers the questions in between.

Connect AI assistants to your data, safely

In a free initial consultation we will look at which access points make sense for your departments and how the tools should be cut.

Request a consultation

Still have questions? We have the answer.

For MCP there is a free sandbox to try things out and an open consultation hour. You will find the dates and the technical description in the help centre.

Frequently asked questions about the MCP server

Can projectfacts be connected to AI assistants such as Claude?
Yes. projectfacts provides an MCP server. Through this open standard you connect AI assistants such as Claude or ChatGPT: the assistant queries data and, within the rights granted, triggers actions, for example querying and creating projects, tasks or time bookings.
Does the AI assistant see all the data in the system?
No. An MCP access never extends permissions. When a tool is called, projectfacts checks the user's personal rights profile, exactly as it does when they log in on screen. Two employees at the same tool therefore see different data.
How is the MCP access set up?
An administrator creates an MCP configuration, selects the tools it contains and releases it for user groups. After that, every employee enters the address in their AI client. Logging in runs through OAuth, and no passwords are handed on.
Why several MCP access points instead of a single one?
Because an assistant works through the list of available tools before every answer. An access with twelve fitting tools delivers faster and more accurate answers than one with a hundred. The usual set-up is a general read access plus one access each for HR, sales and project work.
Which AI applications are supported?
All those that support the Model Context Protocol, among them Claude and ChatGPT as well as self-hosted agents. As the standard is open, you connect projectfacts once and stay free in your choice of assistant.
What is the difference from the AI features in projectfacts?
The MCP server opens projectfacts outwards for an assistant you choose yourself. The AI features run inside projectfacts at particular points instead, for example when a ticket history is summarised. Both assume that you connect an AI of your choice.