Using the MCP server in ATLAS.ti

Who this article is for

Anyone who has already connected an AI client to ATLAS.ti via MCP and wants to understand what they can do with it, what it costs, how their data is handled, and how it works in a team.

Haven't connected a client yet? Start with Setting up the MCP server in ATLAS.ti.

What you can do with the MCP server

Because your AI assistant is connected to your whole project rather than a document you paste in, you can ask it to:

• Find and review your work — list and open projects, browse documents, codes, memos, networks, quotations, and groups, and read the contents of documents and memos.
• Code and annotate — create codes, apply or remove codes on quotations, set code colors, add comments, and link memos to other items.
• Organize your project — create groups and folders, move items into them, and rename or delete entities, including restructuring a messy codebook.
• Bring in new material — create memos and text documents, import documents and transcripts, and convert documents to transcripts.

Tasks that would otherwise take hours — comparing a theme across all your interviews, reorganizing a sprawling codebook, or sorting documents into groups — can often be done in a single conversation, with the results saved straight back to your project.

Good prompts to start with

• "List all the codes in my current project."
• "Summarize the memos in this project."
• "I have 200 codes — some are very similar. Help me get below 100."
• "Read these interviews, create document groups by age bracket, and sort the documents into them."

Cost: what's free and what isn't

It helps to keep three separate things apart:

• The MCP integration itself is free. It's included with every active ATLAS.ti Desktop subscription, at no extra cost — there's no separate add-on or fee.
• Your AI client is not included. You bring your own — for example a Claude subscription, usage-based fees from an AI provider, or just the computing resources to run a local model on your own hardware. You're responsible for obtaining and maintaining whatever accounts or subscriptions your chosen client requires.
• AI Transcription minutes are a separate cost again. They're billed independently of both the MCP integration and your AI client, whether or not you use MCP at all.

Does using MCP use up my AI client's tokens?

Working through MCP does use your AI client's own tokens or usage allowance, since your assistant is doing the reading and writing — the same as any other conversation with that client.

ATLAS.ti's own built-in AI tools are different: Conversational AI, AI Coding, AI Suggested Codes and AI summaries do not draw on your AI client's tokens at all. They run on ATLAS.ti's own AI services (currently OpenAI's models), included in your ATLAS.ti subscription subject to fair-use limits — entirely separate from whatever AI client you connect via MCP. The one exception is AI Transcription, which is always billed by the minute, regardless of which of these routes you use.

MCP vs. ATLAS.ti's built-in AI tools

Since both can involve AI, it's worth knowing when to reach for which:

In practice, many researchers use both: built-in tools for fast, contained tasks while coding day to day, and an MCP-connected assistant for the bigger, cross-cutting work.

Your data and data ownership

A central idea of the MCP integration is that you choose the AI, and you decide where your data goes.

The MCP server runs locally on your own computer and is protected by a private access key. It's off by default and only runs once you explicitly enable it. Using MCP does not send your projects to ATLAS.ti or Lumivero servers — no ATLAS.ti cloud is involved in MCP at all.

Where your data goes from there depends on the AI client you connect:

In short: your findings stay yours. Everything the assistant does is saved in your ATLAS.ti project — structured, portable, and permanent — not trapped in a chat session.

⚠ "Disable AI" does not turn off the MCP server

ATLAS.ti's Disable AI setting turns off ATLAS.ti's own cloud-based AI features — Conversational AI, AI Coding, and AI summaries. It does not disable the MCP server, which is controlled separately (see Setting up the MCP server in ATLAS.ti).

If your organization disables ATLAS.ti's cloud AI for data-governance reasons, remember that MCP is independent: whether any data leaves your computer depends entirely on which AI client you connect. This matters especially for teams at institutions with data governance or ethics review requirements — check both settings, not just one.

Using MCP with a shared or multi-user license

A common question from research teams: if I use MCP to edit our shared project, will my teammates see those changes on their computers too?

No. This is the single most important thing to understand about MCP and teamwork: using MCP does not change how ATLAS.ti Desktop teamwork works, and Desktop teamwork has never had a live shared project.

Here's the key distinction:

• On ATLAS.ti Desktop (Windows and Mac) — where MCP is available — every team member always works on their own separate local copy of the project. There is no shared file that updates automatically for everyone. Combining work has always meant exporting each person's copy and merging them together afterward.
• On ATLAS.ti Web — the only place with a genuinely live, shared project that multiple people can edit and see updates to in real time — MCP is not available at all.

So you can have real-time shared visibility (Web) or an MCP-connected AI assistant (Desktop) — not both at once. Choosing to use MCP means you're on Desktop, and Desktop's teamwork model — asynchronous, merge-based — applies exactly as it always has.

In practice, this means:

• Each team member enables and connects MCP independently, on their own computer, using their own seat on the license.
• Each person's AI assistant reads and edits that person's local copy of the project — not a single live shared file, and not anyone else's copy.
• Combining everyone's work still follows the normal ATLAS.ti Desktop teamwork process: one master project, distributed to the team, with individual copies merged back afterward. See How to work as a team in ATLAS.ti (Windows, Mac, and Web) for that workflow.

Because an AI assistant can rename, restructure, and delete items, this is a good moment to apply the same team discipline you'd use for manual edits: agree on who is editing what before merging, and review AI-proposed changes before accepting them into the master project.

Undo and safety net

Everything the assistant does is saved into your project like any other action, and ATLAS.ti's undo works the same way here as elsewhere: you can undo unlimited actions, including ones the AI made, for as long as your project stays open.

That undo history is tied to your current working session — it does not persist after you close and reopen the project. Once you've closed a project, changes made in that earlier session (by you or by the AI) can no longer be undone through the interface. Regular project backups remain the safest way to protect your work.

Keeping an audit trail of what the AI changed

Because an AI assistant can make many changes in one conversation, it's good practice — especially for team or supervised projects — to keep a record of what it did. A simple way to do this is to ask your assistant directly, for example:

• "Write a memo summarizing every change you just made to this project."
• "Before we finish, list all the codes, groups, and documents you created or renamed in this session."

Since the assistant can create memos directly in your project, this summary becomes a permanent, reviewable part of your project file — useful for your own records, for supervisors, or for justifying analytical decisions later.

Tested AI clients

The ATLAS.ti MCP integration is built on the Model Context Protocol, an open standard, so you're welcome to try it with your preferred client. Compatibility can vary between clients depending on how each implements the protocol.

The clients below are the ones the ATLAS.ti team has tested and confirmed to work. This list will be updated as more clients are validated — check back here for the latest, or ask support if a client you're interested in isn't listed.

This list is provided for informational purposes and reflects the clients verified at the time of writing. ATLAS.ti support covers the ATLAS.ti MCP server itself; the configuration and behavior of third-party AI clients are outside the scope of ATLAS.ti product support. For the most reliable experience, we recommend Claude Desktop.

Using ChatGPT? ChatGPT currently cannot connect to ATLAS.ti. It only connects to MCP servers hosted publicly on the internet, while ATLAS.ti runs its MCP server locally to keep your data private. This is a limitation of ChatGPT, not of ATLAS.ti.

Your data, privacy, and compliance

For the full picture of how ATLAS.ti handles your data more broadly — including Desktop vs. Web storage, GDPR, and other compliance topics — see Data security and privacy in ATLAS.ti, and ATLAS.ti's Legal Center and Privacy Policy: https://atlasti.com/legal

Common issues and mistakes

• Expecting the AI to edit a live shared project with teammates — Desktop teamwork (with or without MCP) is asynchronous. Each person works on their own copy; combine work through the normal project merge workflow.
• Assuming Disable AI also turns off MCP — it doesn't. Turn off the MCP server separately in ATLAS.ti Options/Settings if you need to disable it.
• Not reviewing AI-made changes before merging with the team — treat the assistant like any capable collaborator: review what it changed, especially renames and deletions, before folding its work into a shared master project.
• Assuming AI Transcription is covered by the MCP integration or your AI client — it isn't. Transcription minutes are billed separately regardless of how you use ATLAS.ti's AI features.
• Losing track of what the AI changed — ask your assistant to summarize its changes in a memo before you close the session; see Keeping an audit trail above.
• Nothing happens when I ask my assistant about my project — make sure ATLAS.ti is open with the relevant project loaded; MCP has no project to work on otherwise.

When to contact support

Contact ATLAS.ti Support if:

• Your AI client connects but can't read or change your project as expected.
• You're unsure whether a particular AI client is supported.
• You have questions about how MCP fits with your institution's data governance or ethics review requirements.
• You believe the MCP server sent or received data other than what's described in this article.

When contacting support, please include:

• Your ATLAS.ti version
• Your operating system
• Which AI client you're using
• A description of what you asked the assistant to do.