AI
25. September 2026
How to connect AI to your data sources without compromising compliance
It is tempting to connect Claude or ChatGPT directly to your data sources, such as CRM. Many systems already have a ready-made connector (a so-called MCP) that you can install with a single click. But that conflicts with your GDPR compliance. We have a solution, though!
Content
Do it properly What is PII How Morningbrain works Direct coupling vs. Morningbrain What it means for you The price of MorningbrainJust because it is easy does not mean it is done properly
It is easier than ever to move data from one source to another. And most software offers MCPs that make the connection between AI and the platform even easier.
We have previously written about how connecting your data sources to Claude or ChatGPT allows you to “chat with your data” – simply asking your AI model a question and receiving an answer based on all your data sources. You can read more about chatting with data in this post.
The problem is that all sensitive or identifiable data (PII) comes along with it.
- Names
- Email addresses
- Order details
- Messages
- Phone numbers of your customers’ employees
And much more is sent directly and unfiltered to the AI provider. Neither we nor your customers have agreed to that.
You also send a lot of data to the US, and your data is used for training and a whole lot of other things that are a big “NoNo” under GDPR.
That is why we built Morningbrain.
What is PII?
PII stands for Personally Identifiable Information. It is anything that can point to a specific person, such as a name, an email address, or a phone number. It is exactly the type of data that fills a CRM and that you are responsible for.
How Morningbrain works
Morningbrain is our company brain: an application that sits between your systems and the AI and runs on our own servers. It connects to CRM, resource planning, time tracking, task management, and tracking. When you ask Claude to fetch, for example, deals from Pipedrive, the request goes through Morningbrain – and three things happen before the answer comes back.
First, Morningbrain checks that you have access to the system you are asking about. You only see your own limited slice of the data. Then it masks the personal data in the response, so Claude never sees it.
Finally, it logs the lookup, so you can always see who accessed which data, and when.
Direct connection or Morningbrain?
The one-click connector gets you up and running faster. But you pay for that speed in three ways:
- Everyone in the organisation typically gets the same access level.
- There is no log of who has seen what.
- And personal data leaves the business unfiltered. Traceability in particular is underestimated: in some cases, it is a legal requirement to be able to document who has had access to which data.
Can AI still be useful, then?
Yes, and that is the point. Claude does not need to know the contact person’s name to tell you which deals over DKK 200,000 are stuck in the proposal stage. Company name, deal value, and stage remain intact. Only what points to a specific person is replaced.
When the salesperson needs to call, they look up the name in the CRM, where it belongs.
What does this mean for you?
If your data contains personal data – and data in, for example, a CRM always does – do not start with the one-click connector. Put a layer in between that controls access, masks personal data, and logs lookups. It takes a little longer to get started, but then you can use AI on your sales data without ending up with a compliance problem afterwards.
We set up the same setup for our customers. Get in touch, and we will look at which of your systems makes the most sense to start with.
What does Morningbrain cost?
It depends on which data sources need to be connected, whether the AI should be able to send data to your platforms, whether we need to build workflows, and much more.
For most, it costs somewhere between DKK 25,000–85,000, and then the solution is yours.
In return, you will be 8x faster on selected tasks, with higher accuracy as well.
