You can give Claude Code recurring work such as checking a project folder each morning or preparing a weekly summary. Setting the time is only part of the setup. Whether the task runs on your computer, inside an open session or in the cloud changes what it can do and when it can start.
Three options with different operating conditions
Local Desktop tasks use files and tools on your computer. Each run starts a fresh session, so you do not need to keep an existing conversation open. The Desktop application must remain open and the computer must be awake.
/loop repeats a prompt within an open Claude Code session. It suits temporary checks, such as watching for a deployment to finish. For daily automation that should exist independently of that conversation, choose Desktop or cloud tasks.
Cloud tasks continue when your computer is off. They clone the selected repositories for each run rather than directly using your local folder. That provides a useful decision rule: choose Desktop for work tied to local files, and consider cloud execution when availability must not depend on your machine. Use /loop to keep an eye on work already underway.
Create a local task step by step
The Claude Code desktop experience requires Pro, Max, Ultra, Team or Enterprise. The documented minimum Desktop version for local scheduled tasks is 1.1.5368. Cloud tasks are offered on those plans too, in research preview and subject to organizational permissions. Team and Enterprise owners can disable that access.
- Open Scheduled in the Code tab and select New task. Earlier layouts may call the section Routines.
- Enable Only on this computer for local execution. If a Local and Cloud menu appears, choose Local.
- Give the task a unique name, select the working folder and explain the expected output. If the local form requests a Description, fill that in too.
- Choose the time and recurrence, then the model and permission mode. If requested, approve trust for the selected folder and save.
- Use Run now for an initial run. Open its session through History to inspect the actual output and any pending permission requests.
A sample instruction is: “Review files changed today in this project folder. Report the three most important changes and anything requiring a check. Do not modify files. If it is after 5pm, skip the review and report the late start.” This defines the scope, output and response to a delayed run together.
Sleep, delayed starts and missed runs
A local task can start only while the application is running and the computer is awake. Runs can begin a few minutes after the scheduled time, so do not treat the feature as a second-accurate alarm. Keep computer awake can prevent idle sleep, but closing a laptop lid can still put the machine to sleep.
When the application starts or the computer wakes, Desktop checks for missed runs within the previous seven days. It starts one catch-up run for the most recently missed time and discards older occurrences. It does not replay an entire backlog. A morning task can therefore run at night. For time-sensitive work, include a late-start boundary like the one in the example.
Check permissions and results as well as timing
Each local task has its own permission mode. If a run is waiting for approval, open that session through History and inspect the request. MCP tools requiring user interaction can stop each run even when other permissions have been saved. Granting only the actions the routine needs keeps its scope understandable as it develops.
Start with one folder, a defined date range and one report that is easy to compare with your expectations. This makes the first run useful for checking scope. Deciding early whether you need local file access or machine-independent scheduling is more effective than chasing missed runs afterward.
