How to set up the MCP server in ATLAS.ti

Who this article is for

Anyone using ATLAS.ti Windows or Mac who wants to connect an AI assistant (such as Claude Desktop, Claude Code, or another MCP-compatible client) to their ATLAS.ti project.

If you're looking for what you can actually do once it's connected — including data privacy and working in a team — see the companion article Using the MCP server in ATLAS.ti.

Cost

• The MCP integration itself is included with all ATLAS.ti Desktop subscriptions (Windows and Mac) at no extra cost. There is no separate add-on required to enable it.
• The AI client is not included. ATLAS.ti does not provide or bundle an AI assistant — you bring your own (for example, Claude Desktop, Claude Code, or another MCP-compatible client). Any cost or subscription for that AI client is between you and its provider, not ATLAS.ti — using it with MCP doesn't require any additional purchase from ATLAS.ti.
• Multi-user and campus licenses: if you're part of a multi-user license or a campus license, MCP is included the same way — every user on that license can use the MCP integration without needing a separate purchase.

Before you start

• You need an active ATLAS.ti Desktop subscription (Windows or Mac). MCP is not available in ATLAS.ti Web.
• You need your own AI client. ATLAS.ti does not include or provide one — see the Tested AI Clients section of the companion article for options.
• Mac only: your computer must be running macOS 15 or newer. On older macOS versions, ATLAS.ti's MCP server cannot run, due to a limitation in Apple's Swift compiler that is not expected to be fixed for older macOS releases. If you're on an older Mac, update macOS to version 15 or later before attempting to enable the MCP server.

Step 1: Enable the MCP server in ATLAS.ti

Whichever AI client you plan to use, start by enabling the MCP server in ATLAS.ti itself.

• ATLAS.ti Windows: go to File > Options > ATLAS.ti and enable the MCP server option.

• ATLAS.ti Mac: open Settings and enable the MCP server option.

Step 2: Connect an AI client

There are two ways to connect a client. Use the MCP bundle if you use Claude Desktop; otherwise use the manual HTTP configuration.

Option A: Using the MCP bundle (Claude Desktop)

The MCP bundle (a .mcpb file) is the simplest method — Claude Desktop can install it directly, with no manual configuration needed. This method is currently specific to Claude Desktop.

• Windows: with the MCP server enabled, click Show Bundle in Explorer on the same File > Options > ATLAS.ti page, then copy the bundle file to your C: drive.
• Mac: in Settings, scroll down and double-click the ATLAS.ti.mcpb file.
• In Claude:
  • Go to File > Settings > Extensions > Advanced Settings > Install Extension, navigate to the bundle file, and click Install. Confirm with Install again if prompted.
  • Already have an older bundle installed? In Claude, go to File > Settings > Extensions > Advanced Settings > Install Extension, click Uninstall, then Install again to install the new version.

Option B: Manual HTTP configuration (any other MCP client)

Any MCP client that can connect to an HTTP server can use the ATLAS.ti MCP server this way — for example OpenAI Codex, Goose, or the Gemini CLI. The steps below use Claude Code as a worked example; refer to your own client's documentation for exactly where to add an MCP server.

• Copy the configuration from ATLAS.ti:
  • Windows: go to File > Options > ATLAS.ti and click Copy Configuration.

• Mac: open Settings and click Copy Configuration.

This gives you a small JSON snippet, for example:

{
  "mcpServers": {
    "ATLAS.ti": {
      "headers": {
        "Authorization": "Bearer YFNqUs+T2o5ZZljQsTYrTF6eAhgzj6NQVG0Kf0IZwXg="
      },
      "type": "http",
      "url": "http://127.0.0.1:57000/mcp"
    }
  }
}

• Add it to your client: paste the configuration into your AI client following that client's own documentation for adding an MCP server.
• For Claude Code specifically: install it via its Quickstart if you don't already have it, then add the ATLAS.ti server following Claude Code's 'Connect to MCP servers' instructions.

Step 3: Confirm the connection works

• Make sure ATLAS.ti is open, with a project loaded — the MCP server only has something to work with once a project is open.
• In your AI client, enter a simple prompt such as "List all my ATLAS.ti projects."
• If the connection is working, your client will list your projects. If it doesn't, see Common issues below.

Common issues and mistakes

• ATLAS.ti crashes when I turn on the MCP server (Mac) — this happens on ATLAS.ti 26.1.0 when your Mac is running a macOS version older than 15, due to a Swift compiler limitation on older macOS releases. Update macOS to 15 or later, or update ATLAS.ti to 26.1.1 or later, which blocks the option instead of crashing and explains why.
• I can't enable the MCP server and get a message about updating macOS — this is expected behavior on ATLAS.ti 26.1.1 and later, when your Mac is running a macOS version older than 15. Update macOS to version 15 or later.
• My AI client can't find any projects — make sure ATLAS.ti is open and a project is loaded, not just the app running on the welcome screen.
• I reinstalled the MCP bundle but Claude still uses the old version — uninstall the extension first (File > Settings > Extensions > Advanced Settings > Install Extension > Uninstall), then install the new bundle.
• ChatGPT won't connect — ChatGPT currently only connects to MCP servers hosted publicly on the internet. ATLAS.ti's MCP server runs locally on your computer to keep your data private, so ChatGPT cannot connect to it. This is a limitation of ChatGPT, not of ATLAS.ti.
• My client isn't listed as supported — any MCP client that supports HTTP connections can be used with the manual configuration in Step 2, Option B. Refer to that client's own documentation for where to add an MCP server.

When to contact support

Contact ATLAS.ti Support if:

• The MCP server option is missing from ATLAS.ti Options/Settings.
• You've confirmed you meet the macOS 15+ requirement but still can't enable the MCP server.
• Copying the configuration or the bundle file doesn't work as described.
• Your AI client can't connect even after following the steps for your connection method.

When contacting support, please include:

• Your ATLAS.ti version
• Your operating system and version
• Which AI client you're trying to connect
• A screenshot of any error message.

Note: 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 — refer to that client's own documentation for client-specific issues.