task-started and task-completed read as instructions to the assistant rather than capabilities for you — a call when work begins, a second when it ends with an outcome and details — while auto-task-tracker fills the interval between them, self-triggering every 10 seconds past a 30-second task under a threshold you control. What the tool surface never says is where any of it lands: no tool reads an activity record back, so the value here depends on a destination you would want to confirm before wiring this into anything you rely on.
A task-activity reporting server with 3 tools: a call at the start of a piece of work, a call at the end carrying its outcome and details, and a tracker that reports progress while a long task is still running.
- A start and end marker around any unit of work: task-started records a taskDescription when work begins, and task-completed closes it out with that description plus an outcome and details.
- Progress reported during the run, not only after it: auto-task-tracker is documented as self-triggering every 10 seconds once a task passes 30 seconds, once multiple tools are in use, or when it detects complex processing, and takes a taskThresholdSeconds to move that trigger point.
- Calls aimed at the model rather than at you: task-started and task-completed are both written as instructions to the assistant to call them at the boundary of any task, answer or piece of work, so the reporting happens without a user asking for a status.
No account and no key. The trigger point is the single thing you set: auto-task-tracker takes a taskThresholdSeconds, and task-started and task-completed take the taskDescription you want recorded against the work.
One command — npm install -g agentify-mcp