You can use Countly with Claude Desktop or VS Code + GitHub Copilot so your AI assistant can work with real analytics, events, crashes, and dashboards.

Once connected, you can ask things like: 

• “Show me top events for my app this week.” 
• “Analyze this code and suggest what I should track.” 
• “Which button placement drives better conversion?”

To get more ideas on what you can do check here.

This article guides you to achieve this by setting up the Countly MCP Server.

Getting Started

You'll need the following to get started:

• A Countly auth token, which you can create via our Token Manager on your Countly server 
  • Check this guide for creating a token
  • For simplest use case configure token for all endpoints, multiple times and never expire.

• You'll need to have Node.js installed (one-time setup) 
  • Check official instructions to download Node.js

Self-hosted Countly Users

If you use self-hosted Countly, we suggest running MCP locally or inside your own infrastructure. This prevents you from sending your Countly URL and token to external services. 

Claude Desktop

Claude Desktop is the easiest way to get started. Follow the step-by-step guide below.

Step 1: Open the MCP configuration

Option A (recommended): 

In Claude Desktop, go to Settings -> Developer -> Local MCP Servers -> click Edit Config:

Option B (manual): open the following file: 

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json

Windows: 

%APPDATA%\Claude\claude_desktop_config.json

Step 2: Adding Countly MCP 

If your config is empty, paste the following: 

{
 "mcpServers": {
   "countly": {
     "command": "npx",
     "args": ["-y", "--prefer-online", "countly-mcp-server@latest"],
     "env": {
       "COUNTLY_SERVER_URL": "https://your-countly-url.com",
       "COUNTLY_AUTH_TOKEN": "YOUR_TOKEN"
     }
   }
 }
}

If you already have other MCP Servers, just add Countly inside the existing mcpServers object. For example:

{
 "mcpServers": {
   "existing-server": {
     "command": "..."
   },
   "countly": {
     "command": "npx",
     "args": ["-y", "--prefer-online", "countly-mcp-server@latest"],
     "env": {
       "COUNTLY_SERVER_URL": "https://your-countly-url.com",
       "COUNTLY_AUTH_TOKEN": "YOUR_TOKEN"
     }
   }
 }
}

Step 3: Restart Claude Desktop

Once reopened, you should see the Countly MCP server now listed, as shown in the screenshot below:

Step 4: Verify it works

Open a new (or existing) chat, click on the ➕ (plus) icon, and you should see Countly tools listed, like in the screenshot below: 

You can now try asking questions like "List my Countly apps".  If you get a response, it means the Countly MCP Server was successfully set up!

VS Code + GitHub Copilot

Step 1: Open the MCP config file

Open VS Code and open the Command Palette, select Show and Run Commands:

Run: MCP: Open User Configuration, like in the example below: 

This will open the mcp.json file, where you can add the Countly config described in the next step.

Step 2: Add Countly config

{
 "servers": {
   "countly": {
     "type": "stdio",
     "command": "npx",
     "args": ["-y", "--prefer-online", "countly-mcp-server@latest"],
     "env": {
       "COUNTLY_SERVER_URL": "https://your-countly-url.com",
       "COUNTLY_AUTH_TOKEN": "YOUR_TOKEN"
     }
   }
 }
}

Step 3: Verify it works

Verify that the configuration was set completely. 

Run: MCP: Show Installed Servers from your Command Palette:

You should see "Countly" installed, like in the screenshot below:

Run: MCP: List Servers from your Command Palette to check the Countly server's status: 

If you see that the countly server is "stopped": 

Click on countly and "Start Server": 

Once started, in your VS Code's output console, you should see the logs similar to the following: 

2026-04-23 15:11:28.252 [info] Starting server countly
2026-04-23 15:11:28.253 [info] Connection state: Starting
2026-04-23 15:11:28.254 [info] Starting server from LocalProcess extension host
2026-04-23 15:11:28.254 [info] Connection state: Starting
2026-04-23 15:11:28.254 [info] Connection state: Running

Now that the Countly MCP server is running, you can try @countly list my apps on your GitHub Copilot chat inside VS Code: 

Advanced Configuration

By default, the MCP server exposes the tools available for your Countly instance and token permissions.

You can add extra restrictions to the MCP server configuration using COUNTLY_TOOLS_* environment variables.

This is useful when you want the AI assistant to: 

• only read data
• create notes or alerts, but do not delete anything
• access analytics, but not users or database tools
• use broader token permissions, while limiting what MCP exposes

CRUD Permission Values

You can change the CRUD permissions the AI can perform. 

For example, you can add "COUNTLY_TOOL_ALL": "R" to allow read-only access. 

You'll need to add this under the env section on the MCP's config file:

{
"servers": {
  "countly": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "--prefer-online", "countly-mcp-server@latest"],
    "env": {
      "COUNTLY_SERVER_URL": "https://your-countly-url.com",
      "COUNTLY_AUTH_TOKEN": "YOUR_TOKEN",
      "COUNTLY_TOOL_ALL": "R"
    }
  }
}
}

Enable Selected Categories Only 

For example, you can allow analytics and events as read-only, but disable everything else:

"COUNTLY_TOOLS_ALL": "NONE",
"COUNTLY_TOOLS_ANALYTICS": "R",
"COUNTLY_TOOLS_EVENTS": "R",
"COUNTLY_TOOLS_APPS": "R"

This is a good setup for users who only want to explore analytics.

Read-only Analytics Setup

This allows general read-only access while explicitly disabling more sensitive areas.

"COUNTLY_TOOLS_ALL": "R",
"COUNTLY_TOOLS_DATABASE": "NONE",
"COUNTLY_TOOLS_DASHBOARD_USERS": "NONE",
"COUNTLY_TOOLS_APP_USERS": "NONE"

Allow Limited Write Actions

For example, you can allow your AI assistant to create and read notes, but not update or delete them: 

"COUNTLY_TOOLS_ALL": "R",
"COUNTLY_TOOLS_NOTES": "CR"

Use this when you want AI to help document findings without modifying existing notes.

Allow Crash Workflow Actions

This keeps most tools read-only, while allowing crash-related create/read/update actions if supported by your Countly permissions. 

For example:

"COUNTLY_TOOLS_ALL": "R",
"COUNTLY_TOOLS_CRASHES": "CRU"

Token Permissions Still Matter

It's important to note that MCP restrictions do not override Countly token permissions.

The final access is the more restrictive combination of:

1. What your Countly token allows
2. What MCP server configuration allows

For example: 

• The Auth Token does not allow delete 
• But MCP config says CRUD

As a result, delete will still not work.

Troubleshooting

Security Notes

• Tokens are stored in plain text in your MCP client's config file; treat the file like a secret.
• Don't commit your config to git. Add it to .gitignore if it lives inside a repo.
• Use a dedicated MCP token created in the Token Manager. Don't reuse a personal API key or a token with broad admin scopes.
• Revoke the token in the Token Manager if it's ever exposed, lost, or no longer needed.

What can you do with Countly MCP?