The first question in almost every conversation about AI and project plans is: what happens to my data? The answer has two parts. One concerns the MCP server in Merlin Project, which we built ourselves and can therefore describe precisely. The other concerns the AI provider that receives the server's results. This article explains both parts without sugarcoating and shows what you should clarify with your data protection officer.
The video for this article
The video (in German) is embedded in the German version of this article.
Three properties of the Merlin server
It runs locally. The MCP server is a small program on your Mac. Merlin Project processes every request itself, directly on your computer. There is no Merlin server in the cloud that your data passes through.
It cannot be reached from outside. The server has no web address and opens no access from the network. Only an AI assistant running on the same Mac can talk to it. That is exactly why ChatGPT cannot use the server at the moment. Read more in the article Connecting Merlin Project to ChatGPT?.
It only reads. Through the server, an AI can read tasks, dates, resources, and assignments. It cannot change, create, or delete anything through it.
You decide per project
Even when the MCP server is installed and running, the AI initially sees not a single project. You share each project individually:
- Open the project in Merlin Project.
- Click “AI Tools” in the toolbar.
- Check “Allow AI tools to interact with document”.
When sharing is active, the symbol appears filled. As long as an AI assistant is connected to the project, Merlin Project also highlights the symbol in blue and names the AI accessing it. So you always see whether an AI is reading along right now. Remove the checkmark, and the project is locked again immediately.
What goes to the AI
This is worth a close look, because this is where data leaves your Mac.
If you ask Claude, for example, a question about your project, the AI decides on its own which data it needs for it and queries that data via the server. The server delivers exactly the requested objects, such as the tasks of a phase or the dependencies of a milestone, and not the whole plan automatically. The results of these queries go to Claude and thus to Anthropic.
You should know two points here:
- The AI often makes several queries for one question. If it does not find what it is looking for in one place, it keeps searching in another. In doing so, it may also transfer data that plays no role in the final answer.
- You do not set individually which fields the AI queries. Your control lies in sharing: in principle, the AI can retrieve any information from a shared project during a request.
This leads to the most important recommendation of this article: share only projects whose content you would entrust to the AI provider. If a plan contains particularly sensitive information, such as personnel data in notes, share a cleaned-up copy of the project instead.
What the AI provider does with the data
How the AI provider processes and stores the transferred data is governed by its own terms. Many providers distinguish between personal and business plans here, for example on the question of whether they may use chats to train their models. Check your account settings and the provider's privacy notice before you use project data. If no data at all should go to a provider, use a local language model; the article Connecting Merlin Project to LM Studio shows how.
Quick check: may my project go to the AI?
Whether you should share a project depends on its content. Answer five questions about this project. You get an assessment and a list of what to take care of before sharing:
Does the project contain names of people, for example as resources or in notes?
Does the project contain particularly sensitive information, such as salaries, performance reviews, or health information?
Does the project contain confidential client information, for example under a non-disclosure agreement?
Do you use a plan in which the AI provider does not use your chats for training?
Has your data protection officer approved the AI provider and the plan for your company?
Our assessment
The check is a guide and does not replace legal advice.
Checklist for the data protection officer
| Question | Answer for the Merlin server |
|---|---|
| Where does the server run? | Locally on the user's Mac |
| Can it be reached from the network? | No |
| Can the AI change or delete data? | No, the server only reads |
| Which projects are visible? | Only projects explicitly shared via “AI Tools” |
| Can sharing be revoked? | Yes, at any time and for each project individually |
| What leaves the Mac? | The results of the queries the AI assistant makes |
| Where to? | To the provider of the AI assistant in use |
That leaves the question of the AI provider open. Clarify with your data protection officer which provider and which plan are approved in your company and whether a data processing agreement is in place. This article is a technical description and does not replace legal advice.
The detailed checklist to print and tick off is available as a PDF (in German):
Download the checklist for data protection and IT (PDF, German)
Further reading
The article Which AI client is right for you? describes how to set up the individual AI assistants. The technical details about the server are in the MCP documentation. What MCP is in the first place is explained in the article What is MCP? Explained simply.
If you have any questions about this blog article or would like to discuss it, we look forward to your contribution in our forum.
Frequently asked questions
Does the Merlin Project MCP server run in the cloud?
No. The server is a small program on your Mac. Merlin Project processes every request itself, directly on your computer. There is no Merlin server in the cloud.
Does the AI automatically see all my projects?
No. The AI sees only projects that you have explicitly shared via “AI Tools”. Without sharing, every project stays invisible to it.
Can the AI change or delete my project plan?
No. The MCP server only reads. Through it, an AI can neither change nor create nor delete anything.
How can I tell whether an AI is accessing my project right now?
When the project is shared, the “AI Tools” symbol appears filled. As long as an AI assistant is connected, Merlin Project also highlights it in blue.
Does the AI transfer my entire project plan?
Not automatically. The AI queries exactly the data it needs for your question. For that, however, it often makes several queries and may also transfer data that plays no role in the answer. In principle, any information from a shared project can be transferred.
Who receives the queried data?
The provider of the AI assistant you use, so with Claude, Anthropic. How the provider processes and stores the data is governed by its own terms.
Is my project data used to train the AI?
That depends on the provider and your plan. Many providers handle this differently for personal and business plans. Check your account settings and the provider's privacy notice.
Is using MCP with Merlin Project GDPR-compliant?
The Merlin server itself transfers no data to the outside. Whether passing data on to the AI provider is permitted depends on the provider, the plan, and the content of your project. Clarify this with your data protection officer, for example whether a data processing agreement is in place.
Can I revoke sharing again?
Yes, at any time and for each project individually. Uncheck “Allow AI tools to interact with document”, and the project is locked again immediately.
Why can ChatGPT not use the MCP server?
The server cannot be reached from the network. Only an AI assistant running on the same Mac can talk to it. ChatGPT does not currently support this kind of local connection.
How does my data stay entirely on my Mac?
Use a local language model, for example with LM Studio. The AI then processes the queries directly on your Mac, and no project data goes to an external provider.
