Skip to main content
Scheduling is for work that should run without an open thread. Do not start with a schedule. Start with a manual task, run it once, read the timeline, and make sure the result is useful.
If that run is predictable, then add a schedule.

​
Choosing a schedule

Use schedules for repeatable work such as a morning summary, weekday project check, weekly research digest, or monthly maintenance report. Be careful with tasks that write files, send messages, delete data, rotate credentials, or run broad shell commands. Those should stay manual until you have tested them several times.

​
Time zones

Be explicit about time zones. Your desktop computer, the gateway host, and the people reading the output may not be in the same place. If the UI or gateway build lets you choose a time zone, choose the one that matches the meaning of the task. A workday check should usually follow the work team’s time zone, not whatever time zone the server happens to use.

​
Watch the first runs

After scheduling, inspect the next few runs. Check their start times, gateway, tool behavior, and delivery destination. Pause the schedule when provider credentials change, MCP servers are unstable, task prompts are being edited, or the target system is under maintenance. Resume only after a successful manual run.

​
Further reading