A scheduled task is a saved request that Claude runs for you automatically, again and again, at times you choose.

Anthropic’s Help Center describes scheduled tasks as a way to delegate work to Claude by creating tasks that run automatically on a recurring basis, or on demand. When you create one, Claude saves your prompt as the task’s instructions and runs those instructions at the cadence you choose. In simple words: you write the request once, pick a schedule such as every weekday morning or every Friday afternoon, and Claude takes care of the repeating.

This is a step beyond normal chatting. In a normal chat, you ask and Claude answers right now, while you wait. With a scheduled task, you set things up in advance, and Claude works later, even when you are not looking at the screen. That makes it ideal for recurring chores: a daily summary, a weekly report, a regular check of something that changes over time.

flowchart LR
    A["You write the request once"] --> B["You choose when it should run"]
    B --> C["Claude saves it as a scheduled task"]
    C --> D["At each chosen time, Claude does the work"]
    D --> E["You read the result when it is ready"]
        

A few simple words appear again and again in this topic. Here they are in plain English.

They remove repetition and make sure regular jobs actually happen.

Many of the most useful jobs are not one-off. They repeat. A manager needs a weekly status summary. A student wants a Sunday evening study plan. A shop owner wants a Monday overview of orders. Doing these by hand each time wastes effort, and it is easy to forget. Scheduled tasks turn them into something that simply happens.

Good scheduled tasks are repeating jobs with a clear, checkable result.

They are available on paid plans, in different forms depending on where you use Claude.

According to the Help Center, scheduled tasks are available on paid plans: Pro, Max, Team, and Enterprise. In the Claude desktop app you can find a Scheduled area in the sidebar, which opens a page listing your tasks. In newer versions of the app, the Help Center explains that you can also create a scheduled task straight from a conversation, by asking Claude, then reviewing the proposed name, schedule, and instructions and confirming. For organizations, administrators control access to this kind of feature through their settings.

Claude Code, the product for software engineers, has its own scheduling options. Its documentation describes three ways to schedule recurring work: cloud tasks, desktop tasks, and a lightweight loop command used in a terminal session. They differ in where they run and how long they last.

There are two friendly ways: ask Claude to set it up, or fill in the details yourself.

Way 1: Create it with Claude

The Help Center describes a path where you open the Scheduled page from the sidebar, click to start a new task, and choose the option that lets Claude create it. Claude starts with a prompt asking it to build a scheduled task, you add the details of what you want, and Claude may ask a few multiple-choice questions to fill in the gaps. Finally, it proposes the task, and you confirm by clicking Schedule. The task then appears on your Scheduled tasks page. In newer versions of the app, the same idea works directly from any conversation.

Way 2: Set it up manually

You can also create a task yourself by choosing the manual option and filling in a form. The Help Center lists details such as the task name, the instructions, the schedule, and, in some setups, which folder Claude should work in. Saving the form adds the task to your Scheduled tasks page.

flowchart TD
    A["New scheduled task"] --> B{"How do you want to create it?"}
    B -- "Let Claude set it up" --> C["Describe the job, answer questions, review"]
    B -- "Do it yourself" --> D["Fill in name, instructions, and schedule"]
    C --> E["Confirm with Schedule or Save"]
    D --> E
    E --> F["Task appears on the Scheduled tasks page"]
        

A scheduled task runs without you, so its instructions must be complete and clear.

In a normal chat, if Claude is unsure, it can ask you a question and wait. In a scheduled run, you are usually not there. The prompt is the only guidance Claude has. A vague prompt that works fine in a live chat can produce a muddled result when it runs alone. Think of writing instructions for a capable new colleague who will do the job while you are on holiday and cannot be reached.

Pick a cadence that matches how often you really need the result.

It is tempting to schedule everything to run as often as possible. But each run uses part of your usage allowance, and results you never read are wasted effort. Match the schedule to your real rhythm: a daily task if you act on it daily, a weekly one if you review weekly.

Each run is its own piece of work, with its own record.

When the scheduled time arrives, Claude starts the job using the saved instructions. In the Claude Code documentation, each run creates a new session, where you can later see what Claude did and review any changes. The same idea applies in the apps: a run is like a short conversation that Claude carries out by itself, working through the instructions and producing a result.

Where a task runs decides whether it still works when your computer is off.

Claude Code’s documentation makes the difference clear. Remote or cloud tasks run on Anthropic-managed infrastructure. They keep running even when your computer is off, but they work against a fresh copy of your project rather than your local files. Desktop tasks run on your own machine through the desktop app. They can use your local files and tools, but they only run while the app is open and your computer is awake.

What if your computer was asleep?

The documentation explains that when the desktop app starts or your computer wakes, it checks whether tasks missed runs recently, and may run a catch-up. This is why a task planned for nine in the morning could run much later if your computer slept all day. Each task’s history shows runs that were skipped, so you can see what happened.

flowchart TD
    A["Which kind of task do I need?"] --> B{"Must it run when my computer is off?"}
    B -- "Yes" --> C["Use a cloud task"]
    B -- "No" --> D{"Does it need my local files or tools?"}
    D -- "Yes" --> E["Use a desktop task"]
    D -- "No" --> F["Either can work, so choose the simpler one"]
        

Tasks become much more useful when they can reach your real information.

A scheduled task that only knows what is written in its prompt can do little. The real value comes when it can read from your calendar, your documents, your project tracker, or a folder of files. The Claude Code documentation notes that scheduled tasks can use connected MCP connectors to read from and write to external services during each run. For example, a task might read a team’s chat channel and create entries in a tracking tool.

In the app’s manual setup, the Help Center also lists an optional setting for which folder Claude should work in. Giving a task a specific folder, instead of broad access, limits what it can see and change.

A task that runs by itself must be given clear boundaries.

When you are chatting with Claude, you can watch each step and approve actions as they come up. A scheduled task runs without you, so decisions about what it may do must be made in advance. The Claude Code documentation says each desktop task has its own permission mode, set when you create or edit it, and that you can review and revoke saved approvals later. The principle applies everywhere: decide up front what is allowed.

Once tasks exist, you can look at them, change them, pause them, and remove them.

The Help Center describes a Scheduled tasks page where your tasks are listed, and where you can open any task to see and change its details. The Claude Code documentation lists actions such as editing the prompt, frequency, folder, or other settings, reviewing the history of past runs, reviewing allowed permissions, and deleting the task. It also notes that you can manage tasks by simply asking Claude in a session, for example asking it to show your scheduled tasks or to pause one.

Scheduled runs are real work, so they use real allowance.

A scheduled run is like a conversation that Claude carries out on its own, and it generally draws on your usage in a similar way to ordinary use. A task that reads many documents, searches the web, or runs a long multi-step job can use a lot. If you schedule many tasks, or tasks that run very often, you can reach your usage limits sooner than you expect, and this can even leave less allowance for your normal chats.

Let us follow an imagined project coordinator named Anita as she builds her first scheduled task.

Anita tested the prompt by hand first, kept the task read-only, wrote limits into the instructions, checked the early runs, and keeps reviewing. That is the whole recipe for a reliable scheduled task.

Most problems with scheduled tasks come from a few avoidable habits.

Scheduled tasks save effort, with a few things to manage.

Keep these habits in mind whenever you schedule work.

Quick answers to the questions beginners ask most often.

A saved prompt that Claude runs automatically on a recurring schedule, or on demand, so you do not have to type the same request each time.

According to the Help Center, scheduled tasks are available on paid plans: Pro, Max, Team, and Enterprise. Organizations may control access through admin settings, so check the current article.

Open the Scheduled area in the sidebar, start a new task, and either let Claude set it up by describing the job or fill in the details manually. In newer versions you can also create one from a conversation and confirm with Schedule.

It depends on the kind of task. Claude Code’s cloud tasks keep running when your computer is off, while desktop tasks only run while the app is open and the computer is awake.

Runs are unattended, so you should not count on it. Write the instructions so that Claude has everything it needs, including what to do if information is missing.

Generally each run starts a fresh session. If you need continuity, point the task at a notes file or document that it can read and update.

Yes, through connectors you have set up, with your permission. Give each task only the access it needs.

Yes. Runs are real work and draw on your allowance, so keep tasks focused and match the frequency to how often you read the results.

Start with read-only tasks, state clearly what is forbidden, keep access narrow, and avoid unattended purchases, deletions, or messages. Review early runs carefully.

Open it from the Scheduled tasks page to edit, pause, review history, or delete it. You can also ask Claude in a session to show, pause, or delete your tasks.

Check the task’s history. For desktop tasks, a sleeping computer or a closed app can cause skipped or late runs. Use a cloud task if timing matters.

In the Claude Help Center article on scheduling recurring tasks, and in the Claude Code documentation pages on scheduled tasks for cloud, desktop, and terminal options.

A scheduled task is a saved request that Claude runs automatically at times you choose, so repeating jobs such as a morning briefing, a weekly report, or a regular check happen without you typing the request again. Scheduled tasks are available on paid plans, and you can create one by asking Claude to set it up or by filling in the details yourself. Because runs are usually unattended, the instructions must be complete, with the goal, sources, format, limits, and what to do when information is missing. Cloud tasks keep running when your computer is off, while desktop tasks need the app open and the computer awake. Tasks can use connectors and folders, but they should start read-only, with narrow access and clear limits, because mistakes and misleading text can cause harm when no one is watching. Test your prompt by hand, watch the early runs, match the schedule to your real needs, and review your tasks regularly.