MCP server and GDPR: where does data from the NIS2 walkthrough go?
What a connected AI assistant reads from your walkthrough goes to that assistant's vendor, under your company's agreement with them. The MCP server itself calls no AI vendor. If you want the data to stay in your own network, run the instance, the assistant and the model yourself.
Where does the data go when an assistant reads the walkthrough?
The MCP server is the interface through which an AI assistant reads and fills in your walkthrough, at /api/mcp on nisd2.eu or on your own installation. What the assistant reads comes back as the answer, and it processes that at its vendor, because that is where the language model runs. The MCP server itself sends nothing to an AI vendor.
With Claude on the web or desktop, ChatGPT and Copilot Studio, the vendor's cloud fetches the data. With Claude Code or Cursor, the program on the person's computer fetches it and passes it on to the model it uses. Either way it ends up with the vendor your company chose.
The data in your walkthrough stays where it is. What is new is the route to your assistant's vendor, and a person opens it only once an admin has allowed assistants for your company.
How an assistant can help with your GDPR duties themselves, such as the record of processing activities, is covered in our article on GDPR with an AI assistant and MCP.
Which personal data does the assistant see?
The walkthrough holds names and contact details in a few places: contacts with name and email address on your supplier list (5.1), names and roles of trainees (1.1 and 8.2), and the attendees of the management review (7.3). An assistant that reads one of these items reads those details too.
An assistant sees the content of the areas assigned to the person who connects it. If you want an assistant to see less, have it connected by someone who is assigned only the areas that fit.
Recital 51 of the NIS 2 Directive encourages using AI against cyberattacks and attaches a condition: its use should comply with Union data protection law. It names the principles of data accuracy, data minimisation, fairness and transparency, and data security.
Recital 51 NIS 2
Which agreement do you need with the assistant's vendor?
If the vendor processes personal data on your behalf, it is your processor (Art. 4(8) GDPR). Art. 28 GDPR then asks two things: the processor provides sufficient guarantees for protecting the data, and the processing is governed by a contract or other legal act, usually a data processing agreement (DPA).
Your company concludes that agreement with the vendor itself. nisd2.eu is not a party to it.
For its commercial products such as Claude for Work and the Anthropic API, Anthropic says it acts as a processor on behalf of the customer. By its own account, Anthropic does not use inputs or outputs of these products to train its models by default. Personal accounts are not covered by that statement; connect company data through a business account for that reason.
We have not checked OpenAI's or Microsoft's terms for this page. For each vendor, read its DPA and its rules on training before you connect.
Art. 28 GDPR
How does the data stay in your own network?
If the assistant runs with a language model on your own hardware, what it reads goes to no AI vendor. If the instance is also self-hosted with you, none of it leaves your network. With nisd2.eu and a local model, the data goes from nisd2.eu to you and no further.
This needs a client on your computers that speaks MCP over HTTP and can use a local model. Claude on the web, ChatGPT and Copilot Studio fetch the data into their vendor's cloud, so they do not fit here.
We have not tested such a local setup against the walkthrough yet.
Do you need a data protection impact assessment?
Whether you need a data protection impact assessment (DPIA) before connecting is not something this page answers. Your data protection officer assesses that.
Ask them before an assistant reads items with employees' names, such as the training records, and when the vendor processes data outside the EU.
Either way, you put your assistant's vendor on the list in 5.1 like any other service provider (Art. 21(2)(d) NIS 2). What to consider there is covered in our article on the AI vendor as a supplier.
This page is not legal advice. It describes how the data flows so that your data protection officer can assess it.
How the walkthrough does it
What the assistant sees, and under which agreement, turns into work at these places in the walkthrough.
Areas: you steer what an assistant reads through the areas of the person who connects it. An admin assigns areas in the app.
1.1 and 8.2: the training records hold employees' names. Agree with your data protection officer whether an assistant should read these items.
5.1 and 5.2: your assistant's vendor gets a line on “Your suppliers”. On “What each one has agreed” you tick whether its contract or the DPA settles security requirements and incident reporting.
9.1 Write the cryptography policy: it says where you encrypt, including on the way across the internet. An assistant's connection to nisd2.eu runs over https; a self-hosted instance you put behind TLS yourself.
Common questions
Is our data used to train an AI model?
Your agreement with the assistant's vendor settles that. Anthropic says it does not use inputs or outputs of its commercial products for training by default. We have not checked other vendors' rules for this page.
Is an MCP server a processor under the GDPR?
The MCP server is part of the app that stores your walkthrough anyway, on nisd2.eu or on your own instance, and it calls no AI vendor. What is new is the assistant's vendor: if it processes personal data on your behalf, it is your processor (Art. 4(8) GDPR).
Does the assistant see everything in the walkthrough?
It sees the list of all items with their headline and state. It reads an item's content in the areas assigned to the person who connected it. For an admin that is all of them.
Does nisd2.eu store the conversations with the assistant?
No. The MCP server receives the assistant's calls with what it wants to read or enter. It does not see the conversation between you and the assistant. The server log gets one line per call with the tool and the person's id.
What happens to data already read once we disconnect the assistant?
From its next call it reads nothing more. What it read before is with the vendor. Whether and when it is deleted there is settled by the vendor's terms and your agreement with it.
Read next
Sources
- Directive (EU) 2022/2555 (NIS 2), Recital 51 and Art. 21(2)(d), publications.europa.eu
- Regulation (EU) 2016/679 (GDPR), Art. 4(8) and Art. 28, publications.europa.eu
- Anthropic Privacy Center, Is my data used for model training?, privacy.claude.com
- Anthropic Privacy Center, article 9267385 on its role as processor, privacy.claude.com
- open-isms, lib/mcp/server.ts, lib/mcp/tools.ts and content/docs/self-hosting/mcp.md, github.com/NISD2/open-isms
In practice
Where this comes up in the walkthrough
The walkthrough turns this into steps you fill in, with the law explained beside each one. Open a step to see its screen, filled in for a made-up company.
Supply Chain Security
Write down who works for you
A list of your suppliers and service providers.
Cryptography & Encryption
Write the cryptography policy
Where you encrypt, and who keeps the keys.
What the walkthrough costs
€4,800 net a year, plus VAT
30 days money backBy invoice, due in 30 days. Cancel within the first 30 days: invoice cancelled or money back.
Guided NIS 2 implementation, step by step
A history of changes and sign-offs to show your regulator
Deadlines and reminders for each requirement
Security questionnaire for your suppliers
Export to PDF or CSV at any time
Been asked to handle NIS2?
In the end you have everything in one document: what is done, what is still open, and your management's approval with name and date. The app writes the texts.
