Claude Cowork Scheduled Tasks: Automate Your Morning Routine

Every morning while I drink my coffee, yesterday's meeting notes get summarized and my Linear standup writes itself, and I touch none of it. Scheduled tasks are tasks in Claude Cowork that run on a timer: hourly, daily, weekdays, weekly, or manually. They can use everything Cowork can use, MCPs, connectors, skills, local folders, which is what makes them more than a cron job. One caveat before you plan around them: they only run while your computer is on, so nothing runs overnight. Maybe that changes; today that is where it stands.

How do you create a scheduled task?

Scheduled tasks live in the Cowork tab of the Claude desktop app; as far as I know they are desktop-only. The first one should be a hello world so you see the moving parts:

1. Give it an expressive name and a description, which is effectively the short version of the prompt.
2. Write the prompt itself. For hello world: "say hello world."
3. Pick a model. For simple tasks pick something cheap.
4. Optionally point it at a folder. This is the underrated part: the task can read and create files there, so local strategy docs or notes become injectable context.
5. Pick the schedule: manually, hourly, daily, weekdays, weekly.

The task is saved as a skill file, which is just your prompt turned into a skill, and you can open it to see exactly what runs. Test with a manual run, and edit the instructions whenever the output is off. If you have the iOS app, you get a notification when a run finishes.

Real task: a Linear standup that writes itself

My standup task runs at 9:00. The prompt is nothing special; it asks for an overview of what was resolved yesterday and what is in progress today, and the task pulls the issues over the Linear MCP and writes the report. I used to assemble that picture by hand every morning. It could just as well produce a Word document, a markdown file, or attach itself to the calendar event for the weekly call; inline chat output happens to be enough for me.

Real task: meeting notes with project context

The more interesting task processes my Granola call transcripts. Every morning it reads the transcripts from the last day and writes one markdown file per call, with key takeaways and action items, into a transcripts folder. Two details in the prompt do the heavy lifting:

1. "Do not create two files for the same call": the task checks what already exists, so repeat runs stay idempotent.
2. "Load the right context from the local folder based on the project": my docs folder holds a subfolder per project with strategy, positioning, and roadmap. The task works out which project the call was about and loads that context before evaluating.

That second step is why this beats Granola's own summaries: Granola only knows what was said in the call. This task knows where the project actually stands.

On top of that sits my own MCP, Context OS, which carries around 150 prompts, some of which are lenses: fixed viewpoints like a Lean Startup lens that challenges what it hears. If a transcript sounds like we are about to spend a month on an unvalidated feature, the summary says so. An evaluation needs three things: where we stand, what is new, and a viewpoint. The folder gives the first, the transcript the second, the lens the third.

Grow the automation one decision at a time

The pattern that makes this compound: start with a small workflow and watch what you still do manually after each run. I kept moving the important summaries into Notion by hand, so that became a candidate for the prompt: let the task reason about strategic value and move the keepers itself. It will get that judgment wrong sometimes. The point is that every recurring manual step is a sentence you can add to the prompt, so the workflow absorbs more of the job each week.

Scheduled tasks are the act-on-it counterpart to Live Artifacts, which are the glance-at-it dashboards; that piece covers when each one fits.