What Are Claude Code Scheduled Tasks? Your AI That Works While You Sleep

Wait — What Hit #5 on Hacker News Today?

On March 27, 2026, Anthropic quietly shipped a feature called web scheduled tasks for Claude Code. Within hours it was sitting at #5 on Hacker News with 216 points and 176 comments. Developers were excited. Confused. Asking the same questions over and over.

The short version: you can now tell Claude Code "hey, every morning at 9am, go review my open pull requests and summarize what needs attention" — and it actually does it. While you sleep. Without your computer on. On Anthropic's cloud servers.

That's new. That's a big deal. Let's break down exactly what it is and what it means for how you build.

What Is a Scheduled Task, in Plain English?

Think about it like setting up a recurring calendar appointment — except instead of reminding you to do something, Claude does the thing for you.

You write a prompt. You pick a schedule. Claude runs that prompt on repeat, on Anthropic's servers, and reports back what it did.

Here's a real example of what that looks like in practice:

1. You create a task called "Morning PR Review"
2. Your prompt says: "Check all open pull requests in this repo. For each one, summarize what it does, flag any obvious issues with the code, and note if it's been waiting more than 48 hours for review."
3. You set it to run daily at 8:45am
4. Every morning, Claude clones your repo, reads every open PR, writes a summary, and saves it as a new session you can check over coffee

You didn't write a script. You didn't set up a cron job. You didn't configure a GitHub Action. You just... told Claude what to do and when to do it.

Three Ways to Schedule Claude — Which One Do You Need?

Claude Code actually has three different scheduling systems. Most of the HN buzz is about the new cloud option, but it helps to know how all three compare so you pick the right tool for the job.

	Cloud Tasks ⭐ New	Desktop Tasks	/loop
Runs on	Anthropic's cloud	Your machine	Your machine
Works when laptop is off?	✅ Yes	❌ No	❌ No
Needs open session?	❌ No	❌ No	✅ Yes
Survives restart?	✅ Yes	✅ Yes	❌ No (session only)
Access to local files?	❌ No (fresh clone)	✅ Yes	✅ Yes
Permission prompts?	❌ None (autonomous)	Configurable	Inherits from session
Minimum interval	1 hour	1 minute	1 minute

The rule of thumb:

• Cloud tasks — use when you want it to run reliably without your machine. PR reviews, overnight analysis, weekly audits.
• Desktop tasks — use when Claude needs to touch local files, run local scripts, or use tools only on your machine.
• /loop — use during an active session when you want Claude to keep checking something while you're working. Temporary, disappears when the session ends.

Cloud Tasks: The New Hotness (Deep Dive)

Since cloud tasks are what's generating all the excitement, let's dig into exactly how they work.

Every Run Starts Fresh

When your cloud task fires, here's what actually happens behind the scenes:

1. Anthropic spins up a clean cloud environment
2. Claude clones your GitHub repository (starting from the default branch)
3. Your setup script runs (if you configured one) — things like npm install or pip install -r requirements.txt
4. Claude runs your prompt with access to that fresh codebase
5. If Claude makes changes, it creates a new branch prefixed with claude/
6. The run appears as a new session in your Claude Code session list

There's no memory from the last run. No accumulated state. Each execution is isolated and clean.

It Runs Autonomously — No Permission Prompts

This is the part that makes cloud tasks different from running Claude yourself. When a cloud task fires, Claude runs completely on its own. There's nobody watching. Nobody to approve file changes. Nobody to confirm "yes, go ahead and create that branch."

That means your prompt has to be really clear about what Claude should and shouldn't do. You're writing instructions for someone who will follow them to the letter without asking for clarification.

Claude Makes Branches, Not Commits to Main

By default, when Claude makes code changes during a scheduled task, it creates a new branch named something like claude/pr-review-2026-03-27. It doesn't push directly to main or merge anything on its own.

You stay in control of what actually lands. Claude does the work; you review and merge.

If you want Claude to be able to push to any branch (including main), there's an "Allow unrestricted branch pushes" toggle per repository in the task settings. Use that carefully.

MCP Connectors: Claude with Superpowers

During a cloud task run, Claude can also use your connected MCP services — things like Slack, Linear, GitHub, or Google Drive. These are set up as "connectors" in Claude Code.

When you create a scheduled task, all your connected MCP connectors are included by default. You can remove any that the task doesn't need.

Practical example: a "morning standup" task that reads your open GitHub issues, checks for any Slack messages tagged as blockers, and posts a summary to your #dev channel — all in one automated run.

Schedule Options: How Often Can It Run?

Cloud tasks have a minimum interval of 1 hour. The preset options are:

Frequency	What It Does	Good For
Hourly	Runs every hour	Monitoring, frequent syncs
Daily	Once per day at a time you pick (default: 9am local)	Morning reviews, daily reports
Weekdays	Daily but skips Saturday and Sunday	Work-day-only tasks
Weekly	Once per week on a day and time you pick	Dependency audits, weekly summaries

One thing to know: tasks may run a few minutes after their scheduled time. Anthropic staggers runs across users so servers don't get hammered at exactly 9:00:00am. The offset is consistent for each task — if your task always runs 3 minutes late, it'll always run 3 minutes late.

How to Create Your First Scheduled Task

There are three ways in. All of them end up at the same place — a configured task that runs automatically. Pick whichever feels most natural.

Option 1: The Web Interface (Easiest)

1. Go to claude.ai/code/scheduled
2. Click New scheduled task
3. Name your task (be descriptive — "Daily PR Review" not "Task 1")
4. Write your prompt — this is the most important part
5. Add your GitHub repository
6. Select or create an environment (more on this below)
7. Pick your schedule frequency and time
8. Review which MCP connectors to include
9. Click Create

After you create it, the task shows up in your scheduled tasks list and runs automatically at the next scheduled time. There's also a Run now button on the task detail page if you want to trigger it immediately to make sure it works.

Option 2: The Desktop App

Open the Claude Code desktop app, navigate to the Schedule page, click New task, and choose New remote task. It opens the same form as the web interface.

If you're already living in the desktop app, this is the most convenient path.

Option 3: The CLI (Most Flexible)

From inside any Claude Code session, type:

/schedule

Claude walks you through the setup conversationally — asking what you want it to do, how often, and which repo to use. It feels like telling a colleague what you need.

You can also skip the conversation and pass a description directly:

/schedule daily PR review at 9am

Claude will confirm the details and create the task. If you need to modify a task's schedule later:

/schedule update

Environments: The Thing Most People Miss

When you create a cloud task, you pick an environment. This is where people sometimes get confused, but it's actually pretty simple.

An environment controls what Claude has access to during each run:

• Network access — how much internet access Claude gets during the run
• Environment variables — API keys, tokens, secrets Claude needs (like a GitHub token or Stripe key)
• Setup script — commands that run before Claude starts (like npm install or pip install -r requirements.txt)

There's a Default environment that works out of the box for most tasks. If your task needs specific API keys or a custom setup process, you create a custom environment first, then select it when creating the task.

Real-World Examples You Can Steal

Here are five scheduled tasks worth setting up. These are the kinds of things that eat 30-60 minutes of your week if you're doing them manually.

1. Morning PR Review (Daily, 8:45am Weekdays)

Review all open pull requests in this repository.

For each PR, provide:
- PR title and number
- What it does in 2-3 sentences
- Any obvious code issues, conflicts, or missing tests
- How long it has been open
- Your recommendation: Ready to merge, Needs changes, or Needs review

Format the output as a clean markdown summary. Focus on being useful to the PR author, not exhaustive.

2. Overnight CI Failure Analysis (Daily, 7:00am)

Check the last 24 hours of CI/CD runs in this repository.

For any failed runs:
- Identify which tests or steps failed
- Find the most recent passing run and compare what changed
- Write a 1-paragraph summary of the likely cause for each failure
- If multiple failures share a root cause, group them

Output a brief report I can read in under 5 minutes that tells me where to focus first thing in the morning.

3. Weekly Dependency Audit (Weekly, Monday 7am)

Run a dependency security audit on this project.

Check for:
- Any packages with known CVEs (critical or high severity only)
- Any packages more than 2 major versions behind their current release
- Any packages that haven't been updated in 2+ years

For each issue found, note the package name, current version, recommended version, and a 1-line description of the risk. If no issues are found, say so clearly.

4. Documentation Sync After PR Merges (Daily, 6:00pm)

Check if any pull requests merged today added new API endpoints, changed existing endpoint signatures, or modified database schema.

If yes:
- Update the relevant sections of README.md or docs/ to reflect the changes
- Create a branch called claude/docs-sync-[date] with the updates
- Leave a comment in the PR that triggered the change linking to the docs update

If no relevant PRs merged today, do nothing and note that in your output.

5. Monday Morning Sprint Prep (Weekly, Monday 8:00am)

Prepare a brief sprint kickoff summary for this repository.

Include:
- Open issues labeled "ready" or "sprint" 
- PRs open more than 5 days that still need review
- Any failing CI runs from the past 3 days
- The 3 most recently merged PRs (what shipped last week)

Output as a short Slack-ready message I can paste into #dev. Keep it under 300 words.

What You See After a Task Runs

Every time a scheduled task fires, it creates a new session in your Claude Code session list. You can see exactly what Claude did — every action, every file it read, every branch it created.

This is important: you're not flying blind. Even though the task ran without you, the full work log is there to review. If something went wrong, you can see where Claude got confused or took a wrong turn.

From the task detail page you can also:

• See the history of all past runs
• Click into any run to see the full session
• Hit Run now to trigger an immediate run
• Pause, edit, or delete the task

Who Has Access?

Claude Code scheduled tasks (cloud version) are available to:

• Pro — individual paid plan
• Max — higher usage individual plan
• Team — business plan
• Enterprise — enterprise plan

If you're on the free tier, you don't have access yet. If you're unsure what plan you're on, check at claude.ai/settings.

Claude Code itself is the web and desktop IDE experience at claude.ai/code. If you haven't set that up yet, check out our Claude Code Beginner's Guide first — scheduled tasks are a feature of Claude Code, not the main claude.ai chat interface.

What AI Search Gets Wrong About Scheduled Tasks

Some AI overviews are already getting this slightly wrong. Here's what to watch out for:

"Scheduled tasks can access your local files." — Not cloud tasks. They work from a fresh GitHub clone. Local-only files don't exist in cloud task runs.

"You need to leave Claude Code open." — False for cloud tasks. The whole point is they run without any client open.

"Scheduled tasks run at exactly the time you specify." — Close, but there's typically a small stagger (a few minutes). Plan accordingly if exact timing matters.

"This is the same as the /loop command." — Nope. /loop is session-scoped and disappears when you close Claude Code. Cloud scheduled tasks persist indefinitely until you delete them.

Gotchas to Know Before You Set This Up

Is This Right for You? Quick Checklist

Cloud scheduled tasks are a great fit if you:

• ✅ Have code in a GitHub repository
• ✅ Do the same review/analysis task repeatedly on a schedule
• ✅ Want the task to run even when your machine is off
• ✅ Don't need Claude to touch files that aren't committed to GitHub
• ✅ Are on Pro, Max, Team, or Enterprise

They're probably not the right tool if you:

• ❌ Need Claude to access local files, local databases, or local dev servers
• ❌ Need sub-hourly intervals (use Desktop tasks for that)
• ❌ Want a task that runs as part of an active conversation
• ❌ Are on the free tier

The Bigger Picture: What This Unlocks

Scheduled tasks are part of a bigger shift in how AI coding tools work. Right now, most people use Claude Code interactively — you ask, it answers, you ask again. That's powerful, but it still puts you in the loop for every action.

Scheduled tasks are the first mainstream step toward truly autonomous AI developer assistance. You define the work once, and it happens on its own cadence without you being present.

For vibe coders especially, this is meaningful. You're probably not doing the boring parts of software maintenance because they're tedious and you'd rather be building. Scheduled tasks let you hand those tedious parts off — PR cleanup, documentation drift, dependency hygiene — without writing automation scripts or paying for separate tools.

Think about what you do every Monday morning before you start building. Catching up on what broke over the weekend. Reading through PRs. Checking what's failing in CI. What if all of that was already waiting for you as a clean summary when you opened your laptop?

That's the idea. And it's live today.

If you want to understand the bigger picture of how Claude Code stacks up against other AI coding tools, we did a deep dive in our Most Reliable AI Coding Service guide. And if you're curious how Claude Code usage translates to actual token costs, Claude Code by the Numbers has a breakdown worth bookmarking.

What to Learn Next

• Claude Code Beginner's Guide — If you haven't fully set up Claude Code yet, start here before creating scheduled tasks
• Most Reliable AI Coding Service — See how Claude Code compares to Cursor, Copilot, and Codex for everyday use
• What Is End-to-End Testing? — Once scheduled tasks are reviewing your PRs, you'll want solid test coverage for what they're evaluating
• What Is API Key Management? — Essential reading before you add API keys to your task environments
• Claude Code by the Numbers — Understand how token usage works so scheduled tasks don't surprise you on the bill

FAQ