Time Trakkr runs a Model Context Protocol (MCP) server, so you can point an AI client such as Claude, ChatGPT, Gemini CLI, Perplexity, Claude Code, or Cursor directly at your workspace. From there you can ask about tracked time and budgets, log time, run timers, and draft invoices in plain language — all governed by your role, the same as everywhere else in the app. The connection is set up from the Integrations page, in the Connect your AI (MCP) card.
The card shows your Server URL, which is:
https://timetrakkr.com/api/mcp
Copy that URL; every client below needs it.
Every client below signs in with OAuth: a Time Trakkr sign-in page opens, you approve, and the connection acts as you, with your role.
Open Settings → Connectors → Add custom connector, paste the server URL, and sign in when prompted.
Custom connectors need a paid ChatGPT plan (Plus, Pro, Business, or Enterprise).
Time Trakkr, paste the server URL, and choose OAuth for authentication.The Gemini CLI connects; the Gemini web app does not take custom MCP servers.
Add Time Trakkr to ~/.gemini/settings.json:
{ "mcpServers": { "timetrakkr": { "httpUrl": "https://timetrakkr.com/api/mcp" } } }
Start gemini and run /mcp auth timetrakkr. Your browser opens the Time Trakkr sign-in.
After you approve, the browser shows a plain "Authentication Successful!" page on a localhost address. That page comes from the Gemini CLI on your own computer, which is how it receives the sign-in. Close it and go back to the terminal.
Run /mcp to check that the Time Trakkr tools are listed.
Custom connectors need a paid Perplexity plan.
Time Trakkr, paste the server URL, choose Streamable HTTP, and choose OAuth. Leave the client ID and secret empty; Perplexity registers itself.Or import the Time Trakkr plugin, which adds the same connector for you: open Plugins → Add custom → Import plugin from GitHub, paste https://github.com/time-trakkr/time-trakkr-plugin exactly as written (lowercase), then connect and sign in.
Run claude mcp add --transport http timetrakkr https://timetrakkr.com/api/mcp, then run /mcp to authenticate.
Add the same server URL as a custom MCP server in Cursor's settings and authenticate when prompted.
When you sign in over OAuth, the consent screen lets you choose read-only or read/write access, so you decide up front whether the client may change anything.
There are three credentials that work, and they behave differently:
Bearer header, but over MCP they are read-only and belong to the organization rather than a person. Use one when you want a machine to read data without acting as any individual.Every call is scoped to your workspace and follows the same rules as the rest of Time Trakkr:
Each person can make 100 requests every 15 seconds, the same limit as the REST API; an organization key shares one allowance per workspace. Every message in a batch counts. Past the limit the server answers 429 with a Retry-After header giving the seconds to wait. Normal use of an AI client comes nowhere near it.
status.timetrakkr.com shows whether the app and the MCP server are up, with the history of any incident. It checks the server from outside every few minutes, so it reflects what a client sees. Planned maintenance is posted there in advance.
The server implements MCP protocol version 2025-06-18 and also accepts 2025-03-26 and 2024-11-05; the version is agreed when a client connects. New protocol versions are added beside the ones already supported rather than replacing them.
For working with the built-in assistant inside Time Trakkr rather than an outside client, see the AI assistant. For the plain REST API and webhooks, see the API and webhooks.
Updated 2026-11-09. Still stuck? Contact support or return to the Help Center.