MCP with AI assistants

The usertrax MCP server lets AI assistants query your live analytics (dashboard, conversions, sessions, feedback, surveys). A/B testing via MCP (analyze, create, update, delete) requires the Business plan. Consent covers read and write for those tools.

https://api.usertrax.io/mcp/platform

Same URL in the dashboard under Settings → MCP / AI Assistants.

usertrax uses OAuth 2.1 (browser login with magic link). You do not create an API/MCP token or paste a Bearer header into config — the client opens a browser, you confirm consent (usertrax Platform MCP, Lesen und Schreiben), done.

The token is tied to your user. Pick teams in the conversation (list_teams → team_id); otherwise your current team is used.

1. Open Customize → Connectors (or Settings → Connectors)
2. Add a custom connector
3. URL: https://api.usertrax.io/mcp/platform
4. Connect → magic link + consent in the browser
5. Use the connector in chat and ask analytics questions

1. Enable Developer Mode in ChatGPT settings
2. Connectors (or Plugins) → create a connector
3. MCP URL: https://api.usertrax.io/mcp/platform, auth: OAuth
4. Connect → magic link + consent
5. Enable the connector in the chat

claude mcp add --transport http usertrax https://api.usertrax.io/mcp/platform

Optionally add --scope user for all projects.

Then start Claude Code, run /mcp → usertrax → Authenticate → browser login. Status: claude mcp list.

Native HTTP URL in MCP settings (OAuth with cursor:// callback):

{
  "mcpServers": {
    "usertrax": {
      "url": "https://api.usertrax.io/mcp/platform"
    }
  }
}

On first use: Authenticate → magic link + consent in the browser.

Alternative via mcp-remote (stdio bridge, local http:// callback):

{
  "mcpServers": {
    "usertrax": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.usertrax.io/mcp/platform"]
    }
  }
}

• “Which teams do I have — which ones have MCP?”
• “Dashboard for the last 7 days.”
• “Conversions this week by domain.”
• “Running A/B tests — any significance yet?”
• “New feedback with comments?”

Full tool list: Tool reference.

• Read-only tools only, rate-limited endpoint
• No domain auth_keys, no export credentials
• Plan and team checks on every call
• Consent revocable (remove the connector in the client)