Claude Code Scheduled Tasks: Put Your AI Agent on Autopilot

Claude Code scheduled tasks let you save a prompt once and have it run automatically on a recurring cadence. No more opening a new session every morning to type the same "summarize my Slack threads" prompt. No more forgetting to check error logs before they pile up. You define the work once, Cowork handles the rest.

Here's the fastest way to set one up:

1. Open Claude Desktop's Cowork tab
2. Type /schedule in the task input
3. Describe what you want done and how often
4. Click Schedule to confirm

Claude now runs your prompt at the cadence you choose: hourly, daily, or every weekday morning before you sit down at your desk.

What Are Scheduled Tasks?

Scheduled tasks are a new feature in Claude Code Desktop (launched March 6, 2026) that lets you save a prompt as standing instructions and run it on a recurring cadence. Each run launches as its own independent Cowork session. Think of it as cron for your AI agent, except the agent has access to every tool, MCP server, skill, and plugin you've connected.

This lives in the Cowork tab of the Claude Desktop app. Cowork is the autonomous agent side of Claude Desktop, running Claude Code's agentic architecture for knowledge work beyond pure coding. If you've used Remote Control to manage sessions from your phone, you've already seen pieces of this infrastructure.

The critical thing to understand: a scheduled task isn't running a text generation job. It's launching a full agent session. That agent can browse the web, read emails, query databases, interact with APIs, write to files, and orchestrate multi-step workflows. The scheduled trigger just automates the "start a new session with these instructions" step.

Requirements:

• Claude Desktop app (macOS or Windows only)
• Paid plan: Pro, Max, Team, or Enterprise

How to Set Up Scheduled Tasks

There are two ways to create a scheduled task. The /schedule command is conversational and guided. The Scheduled Tasks page gives you a direct form.

Method 1: The /schedule Command

This is the more natural approach if you're already in a Cowork session.

1. Open the Cowork tab and click "+ New task" (or use an existing task)
2. Type /schedule in the task input
3. Add descriptive details about what you want automated, then click "Let's go"
4. Claude may ask multiple-choice questions to gather specifics (frequency, scope, data sources)
5. Claude outputs: a task name, schedule cadence, and summary of what it will do
6. Click "Schedule" to confirm

The conversational flow is useful when you're not sure exactly how to structure the task. Claude will help you refine the prompt and pick an appropriate frequency.

Method 2: The Scheduled Tasks Page

For direct control without the conversational setup:

1. Click "Scheduled" in the left sidebar
2. Click "+ New task" in the upper right
3. Fill in the form:
   • Task name: A short, descriptive label
   • Description: What the task accomplishes
   • Prompt: The full instructions Claude will follow each run
   • Frequency: Hourly, Daily, Weekly, Weekdays, or Manual
   • Model: Which Claude model to use
   • Working folder: The directory context for file operations
4. Click "Save"

The form method is faster when you already know exactly what you want. It's also easier for duplicating tasks: create one, then modify copies for variations.

Cadence Options

You get five scheduling frequencies:

• Hourly: Runs every hour the app is open
• Daily: Once per day
• Weekly: Once per week
• Weekdays: Monday through Friday only
• Manual: On-demand only, no automatic runs

For most use cases, daily or weekday schedules make the most sense. Hourly is useful for monitoring tasks like error log checking.

Use Cases That Actually Matter

The versatility comes from Cowork running Claude Code under the hood. Any MCP server you've configured, any integration you've connected, any file on your system is available to the scheduled task.

Error Log Monitoring

Thariq from the Claude Code team shared his favorite: "Ask it to check error logs every few hours and create PRs for any actionable errors."

Set this to hourly. Connect your logging service via MCP. The agent reads recent errors, filters noise, and opens pull requests for issues worth fixing. You review PRs instead of scanning logs.

Morning Briefings

Schedule a weekday task that runs before your workday starts. Connect Slack, email, and calendar via MCP servers. The agent summarizes overnight messages, flags urgent items, and outlines your day. You open Claude Desktop to a ready-made briefing instead of spending 30 minutes triaging inboxes.

Weekly Reports

Set a weekly task that pulls data from Google Drive, spreadsheets, or project management tools. The agent compiles metrics, identifies trends, and generates a formatted report. Especially powerful for teams running async workflows where status updates are scattered across multiple tools.

Recurring Research

Track competitors, industry news, or technology trends on a daily or weekly cadence. The agent browses configured sources, extracts relevant updates, and compiles summaries. Connect it to web search and RSS feeds for broad coverage.

File Organization

Periodically sort, clean, or process files in a working directory. Useful for teams that accumulate artifacts from multi-agent task runs: clean up temp files, archive completed outputs, organize by date.

Team Standup Summaries

Connect project management tools (Jira, Linear, GitHub) via MCP and schedule a weekday summary. The agent pulls recent activity, groups by team member, and generates a standup-ready update. No more "what did you do yesterday?" roundtables.

Managing Your Schedules

All scheduled tasks live in the "Scheduled" section of the left sidebar. From there you can:

• View all tasks: See every scheduled task with its cadence and status
• Review runs: Check upcoming and past executions, including results
• Edit instructions: Modify the prompt, frequency, or model at any time
• Pause/resume: Temporarily disable a task without deleting it
• Delete tasks: Remove tasks you no longer need
• Run on demand: Trigger any task immediately, regardless of its schedule

Past runs work like any regular Cowork session. You can open a completed run, review what the agent did, inspect outputs, and follow up with additional prompts if needed.

Limitations and Workarounds

Honest constraints you should know about:

Your computer must be awake and Claude Desktop must be open. This is the biggest limitation. If your machine sleeps or the app closes at the scheduled run time, Cowork skips that run. When the computer wakes or the app reopens, Cowork automatically re-runs any skipped tasks and shows a notification. But if your laptop is closed from Friday evening to Monday morning, a daily task won't run three times on Monday. It re-runs once.

Desktop only. Scheduled tasks are not available on the web or mobile versions of Claude. They're also not available in the Claude Code CLI. This is a Cowork-specific feature tied to the Desktop app.

No cloud execution. Unlike some CI/CD-based approaches, the agent runs locally on your machine. There's no server running your tasks while you're offline. This trades always-on reliability for full local access to your files, tools, and integrations.

Workaround for always-on scheduling: If you need tasks to run regardless of your machine's state, keep a dedicated machine (even a cheap desktop or mini PC) with Claude Desktop running. A Mac Mini or similar always-on device with the app open handles this well.

The OpenClaw Effect

If you've been following the OpenClaw phenomenon, this feature should look familiar. OpenClaw's entire pitch was an autonomous AI agent that connects to your apps and runs tasks on your behalf, on a schedule, without you babysitting it. That pitch resonated with 199K GitHub stars and a wave of mainstream attention that proved the demand was real.

Anthropic clearly took notes. Remote Control landed in February as a direct answer to OpenClaw's "control your computer from your phone" feature. Now scheduled tasks close another gap: the ability to set an agent loose on a recurring cadence without manual intervention.

The pattern is consistent. OpenClaw shows what users want through sheer viral adoption. Anthropic ships the native, integrated, security-first version weeks later. Remote Control replaced OpenClaw's WebSocket-based remote access with an encrypted, outbound-only bridge. Scheduled tasks replace the "always-on personal assistant" concept with a structured, permission-controlled agent scheduler that inherits your full MCP and plugin setup.

The difference in execution matters. OpenClaw grants broad system access by default and has faced security incidents (CVE-2026-25253 affected 50K+ instances). Cowork's scheduled tasks run in an isolated VM with explicit permission boundaries. OpenClaw requires self-hosting, port forwarding, and manual configuration. Cowork gives you a form and a sidebar.

This isn't a criticism of OpenClaw. It proved the category. But for developers already in the Claude ecosystem, Anthropic building these capabilities natively means you get the same autonomous agent patterns with tighter integration, better security, and zero infrastructure overhead.

Scheduled Tasks vs. DIY Cron

Before Anthropic shipped this, the community also built Claude-specific scheduling solutions. Tools like runCLAUDErun (macOS app), claude-code-scheduler (a plugin using launchd, crontab, and Task Scheduler), and claude-tasks (a TUI scheduler with cron) all addressed this gap. Some developers wired up GitHub Actions or raw cron jobs to trigger CLI sessions.

Cowork scheduled tasks is Anthropic's native answer to all of them. Here's how they compare:

Native scheduled tasks win on ease of setup, integrated management, and MCP access. DIY solutions win on always-on reliability and cloud execution. If your tasks can tolerate running only when your machine is awake, the native approach is simpler. If you need 24/7 guaranteed execution, cron or GitHub Actions still have the edge.

Frequently Asked Questions

Do scheduled tasks work in Claude Code CLI? No. Scheduled tasks are exclusive to the Cowork tab in Claude Desktop. The CLI doesn't have a built-in scheduler. For CLI-based scheduling, you'd need cron, Task Scheduler, or a third-party tool like claude-code-scheduler.

What happens if my computer is asleep when a task is scheduled? Cowork skips the run. When your machine wakes or you reopen Claude Desktop, it automatically re-runs the missed task once and sends a notification. It won't run multiple catchup executions for multiple missed windows.

Can I use scheduled tasks on Linux? Not currently. Claude Desktop is available on macOS and Windows only. Linux users can approximate this with cron jobs running claude -p in headless mode.

Do scheduled tasks count against my usage limits? Yes. Each run launches a full Cowork session, which consumes more usage than a regular chat. Factor this in if you're on a Pro plan with tighter limits. Max and Enterprise plans have more headroom.

Can a scheduled task create pull requests? Yes. The agent has full access to your local git setup and any configured MCP servers. If you connect GitHub via MCP, the agent can create branches, commit changes, and open PRs.

What This Means for Development Workflows

Scheduled tasks turn Claude from a tool you invoke into an agent that works alongside you on a cadence. Instead of remembering to run a prompt, you define the intent once and let the system handle timing.

Combined with Remote Control (monitor sessions from your phone) and autonomous agent loops (long-running agentic work), scheduled tasks complete the picture of Claude as a persistent development companion. Remote Control lets you steer sessions from anywhere. Autonomous loops let agents work independently for hours. Scheduled tasks automate the trigger itself.

For teams using multi-agent architectures, this opens a practical pattern: schedule a daily task that reviews overnight agent work, summarizes results, and flags items needing human review. The agent becomes both the worker and the shift supervisor.

Pick one recurring task you do manually today. A morning inbox triage, a weekly metrics pull, an error log scan. Set it up as a scheduled task. Run it for a week. See how much time comes back.

Start with /schedule and build from there.