zaplitny/atimelogger-mcp

This project is a small server that connects the ATimeLogger time tracking app to AI assistants like Claude and OpenAI's Codex, using a standard called MCP, the Model Context Protocol. Once set up, it lets an assistant read and control a person's time tracking data directly through normal conversation, instead of the user opening the ATimeLogger app themselves. To use it, a person first generates a personal access token inside their ATimeLogger account settings, then registers the server with their AI assistant using a one line command or a small configuration file, depending on which assistant they use. No separate install step is required since the tool is fetched automatically when it first runs. It can also run locally on a person's own computer, or be hosted remotely so it can be reached from Claude in a web browser or on mobile, since those versions cannot run a local server directly. The server exposes a set of specific actions the assistant can take: checking what activity is currently running, listing the names of activity types, starting or stopping an activity, pausing and resuming one, logging a completed activity after the fact with a specific start and end time, editing notes or tags on an existing entry, and pulling summary reports or raw history for a chosen time period. A person can refer to activities by their everyday name rather than an internal code, and the assistant will ask for clarification if a name is ambiguous. In practice this means a person can simply say things like start tracking work, I actually began the meeting at two o'clock, log two hours of reading from last night, or ask where did my week go, and the assistant translates that into the correct action or report. The project is written in TypeScript, requires Node.js version 20 or newer to run from source, and is distributed as a published npm package for the simplest setup path.