Use Teetrack’s MCP server from an AI assistant
By Tristan Doehl · · Updated
If you already work in an AI assistant, opening a separate application for every timer change interrupts that workflow. Teetrack’s MCP server lets a compatible assistant read and change your time records through tool calls, while the records remain available in the web application.
What the connection adds
MCP stands for Model Context Protocol, a standard for connecting AI applications to external tools. In Teetrack, those tools cover timers, completed entries, projects, tags, clients, summaries, settings, and calendar review. The assistant selects tools in response to your request and uses their returned data to form an answer.
For example, “Start a timer for checkout testing on Acme redesign” requires the assistant to identify the project and create a running entry. The project name needs to match the intended engagement. If there are several plausible matches, resolving that ambiguity is part of the task; the protocol does not determine which client you meant.
Connect an account and check access
The hosted MCP endpoint is https://teetrack.it/api/mcp. Most assistant applications connect through OAuth: you add the server URL, sign in to Teetrack in your browser, and approve the requested permissions. A personal API token is only needed for clients without OAuth support. To create one, open MCP in the sidebar, expand Advanced: Manual Token under MCP Integration, and choose Generate Token. Copy it when shown and use it in the application’s connection settings. Each account has a single token, so regenerating it disconnects any client that still uses the old one.
Start with a read request, such as listing your projects or running timers. Check the response against the web application before asking the assistant to create an entry. This confirms both that the connection works and that you authorised the intended account.
With OAuth, the consent page lets you untick permission groups, and the assistant can then use only the tools those permissions cover. A manual token has no such limits: it grants every tool, including deletions. Subscription restrictions still apply to operations within those tools. Check the returned result before treating a conversational answer as confirmation that a change succeeded.
Follow a time entry through the tools
Suppose you ask the assistant to start work on a project, then later ask it to stop the timer. The stored entry can be inspected in the dashboard just like one created in the browser. You can also ask for a list of the day’s completed entries and compare their descriptions and durations with the work you remember. If calendar sync is on, ask for completed entries specifically, because an unfiltered list also contains unconfirmed calendar drafts.
Logging a past session needs more detail. Supply the date, start and end times, time zone where relevant, project, and description. Review the created entry, especially if the request used a relative date or crossed midnight. Natural language makes the request convenient, but it does not guarantee that every interpretation is correct.
Understand what leaves the account
Tool responses are returned to the assistant you connected. If that assistant is hosted by another provider, the data in those responses is processed by that provider under its policies and your account settings. Account isolation in Teetrack does not make a guarantee about another service’s retention or training practices.
Check deletion targets before approving them
MCP deletions are permanent. Inspect the matching records and their IDs before authorising a deletion, and use the assistant application’s approval controls to review the proposed action. Correct an incorrect duration by editing the entry when that is sufficient.
Build from a small, checked workflow
Begin with one project and a timer you can inspect. Once that works, the Claude walkthrough, ChatGPT and Gemini guide, and advanced workflows develop the same connection into daily tracking and reporting. A checked result is the foundation for trusting a longer sequence of tool calls.
