Anthropic shipped Claude Routines on April 14, 2026 — and it quietly changes what it means to use an AI coding assistant. Instead of spinning up a session, typing your instructions, and babysitting the output, you define your task once and Claude runs it on its own: on a schedule, via an API call, or in response to a GitHub event.
This guide covers everything you need to know: what Claude Routines actually are, how they differ from cron jobs and GitHub Actions, how to set one up, real use cases, and the plan limits you need to plan around.
What Are Claude Routines?
A Claude Routine is a saved Claude Code configuration — a prompt, one or more GitHub repositories, and a set of connectors — packaged once and executed automatically on Anthropic-managed cloud infrastructure.
The key difference from anything you've used before: your laptop doesn't need to be on. Routines run in Anthropic's cloud, not on your local machine.
"A routine is a saved Claude Code configuration: a prompt, one or more repositories, and a set of connectors, packaged once and run automatically."
— Anthropic documentation
Each routine can be triggered three ways:
| Trigger Type | How It Works |
|---|---|
| Scheduled | Runs on a cadence: hourly, nightly, weekdays, or weekly |
| API | Fires when any system sends an HTTP POST to a per-routine endpoint |
| GitHub Event | Reacts to repo events: PR opened, push, release, issue, and more |
A single routine can combine all three triggers. A PR review routine, for example, could run nightly and react to every new PR and trigger from a deploy script.
Why Claude Routines Matter: The Problem They Solve
Before Routines, developers using Claude Code for automation had to manage their own cron jobs, maintain their own infrastructure, and wire MCP servers together manually. Every automation depended on a terminal staying open.
Routines eliminate that overhead entirely. You write the prompt, connect a repository, pick a trigger, and the automation runs on Anthropic's cloud — no servers to provision, no cron jobs to debug, and no local machine dependency.
The shift is significant: Claude Code moves from a tool you use interactively to something closer to a persistent background service — an AI that keeps working on your codebase while you sleep.
Claude Routines vs. Cron Jobs vs. GitHub Actions
Many developers ask: do I need this if I already have cron jobs or GitHub Actions? Here's the honest comparison:
| Cron Jobs | GitHub Actions | Claude Routines | |
|---|---|---|---|
| What runs | Fixed scripts | Fixed workflows | AI-driven reasoning |
| Adapts to context | ❌ | ❌ | ✅ |
| Cloud-hosted | You manage it | GitHub's infra | Anthropic's infra |
| Triggers | Schedule only | Schedule + GitHub events | Schedule + API + GitHub events |
| Best for | Deterministic tasks | CI/CD pipelines | Judgment-based, recurring tasks |
The core distinction: cron jobs and GitHub Actions run set scripts without dynamic input from an AI model. Claude Routines prompt an AI model on a schedule or trigger and take different actions depending on the context they encounter.
Think of a Claude Routine as a dynamic cron job — or a short-lived, trigger-driven AI agent that spins up, does the work, and exits cleanly.
When to use Routines over GitHub Actions:
Tasks requiring judgment — triage logic, code review checklists, docs-drift detection, cross-language porting. If the task is mechanical and deterministic (build, test, deploy), GitHub Actions is faster and more predictable.
Real-World Claude Routines Use Cases
These are the patterns that have already emerged from early adopters and Anthropic's own documentation:
1. Nightly Backlog Grooming
Set a scheduled routine to run every weeknight. Claude reads issues opened since the last run, applies labels, assigns owners based on the area of code referenced, and posts a summary to Slack. Your team starts the day with a groomed queue — no one had to touch it.
2. Alert Triage on Autopilot
Wire your monitoring tool (Datadog, PagerDuty, etc.) to a routine's API endpoint. When an error threshold is crossed, Claude pulls the stack trace, correlates it with recent commits, and opens a draft PR with a proposed fix and a link back to the alert. On-call reviews a PR instead of starting from a blank terminal.
3. Bespoke Code Review on Every PR
A GitHub trigger fires on pull_request.opened. Claude applies your team's review checklist, leaves inline comments for security, performance, and style issues, and adds a summary comment. Human reviewers can focus on architecture instead of mechanical checks.
4. Deploy Verification
Your CD pipeline calls the routine's API endpoint after each production deploy. Claude runs smoke checks, scans error logs for regressions, and posts a go/no-go summary to the release channel before the deploy window closes.
5. Docs Drift Detection
A weekly schedule trigger scans merged PRs, flags documentation that references changed APIs, and opens update PRs against the docs repository for an editor to review.
6. Cross-Language SDK Porting
A GitHub trigger fires on merged PRs in one SDK repository. Claude ports the change to a parallel SDK in another language and opens a matching PR — keeping two libraries in sync without a human re-implementing each change.
How to Set Up a Claude Routine (Step-by-Step)
Requirements: Claude Code on Pro, Max, Team, or Enterprise plan with Claude Code on the web enabled.
Option A: From the Web UI
- Go to claude.ai/code/routines and click New routine
- Give it a descriptive name and write your prompt — this is the most important step. The routine runs autonomously with no approval prompts, so make the prompt explicit about what to do and what success looks like
- Select one or more GitHub repositories (Claude creates
claude/-prefixed branches for its changes) - Choose a cloud environment (controls network access, environment variables, and setup scripts)
- Add one or more triggers (schedule, API endpoint, or GitHub event — you can combine all three)
- Review connectors — all your connected MCP connectors (Slack, Linear, Google Drive, etc.) are included by default; remove any the routine doesn't need
- Click Create → use Run now to test immediately
Option B: From the CLI
In any Claude Code session, type:
/schedule daily PR review at 9am
Claude walks through the setup conversationally and saves the routine to your account. It appears at claude.ai/code/routines immediately.
Note:
/schedulein the CLI creates scheduled routines only. To add API or GitHub triggers, edit the routine on the web.
Option C: From the Desktop App
Click New task in the sidebar → choose New remote task. (Choosing New local task creates a local Desktop scheduled task that runs on your machine — this is not a routine.)
Writing a Good Routine Prompt
Because routines run autonomously with no permission prompts and no human in the loop, the prompt is everything. A few principles:
- Be explicit about what success looks like. Don't just say "review PRs" — say "review the PR for security issues, performance regressions, and naming convention violations. Leave one inline comment per issue. Add a summary comment at the top."
- Scope the routine tightly. Only include the repositories and connectors it actually needs.
- Include output instructions. Tell Claude where to post results (Slack channel, PR comment, Linear issue).
- Anticipate edge cases. What should Claude do if there's nothing to triage? If a repo is empty?
API Trigger: Wiring Routines Into External Systems
Every routine can expose its own HTTP endpoint with a dedicated bearer token. Once saved, copy the URL and token from the routine's detail page.
You can pass context with every trigger via the text field in the request body — it gets appended to the routine's configured prompt as a one-shot user turn. This is how you pipe alert payloads, deploy metadata, or ticket content directly into Claude's context.
curl -X POST https://api.anthropic.com/v1/claude_code/routines/trig_01ABC.../fire \
-H "Authorization: Bearer YOUR_TOKEN" \
-H "Content-Type: application/json" \
-d '{"text": "Alert: P1 error in payments-service, trace ID 9f2a..."}' Claude opens a session, runs your prompt with the alert as context, and returns a session URL you can open in real time.
Plan Limits and Daily Run Allowances
Routines draw from your subscription's usage limits. There are also daily run limits per plan:
| Plan | Daily Routine Runs |
|---|---|
| Pro | 5 per day |
| Max | 15 per day |
| Team | 25 per day |
| Enterprise | 25 per day |
Extra runs beyond these limits are available via metered overage billing if extra usage is enabled on your account.
Minimum schedule interval: 1 hour. Per-minute polling is not supported by Routines — use /loop inside a session for that.
Current Limitations (Research Preview)
Claude Routines launched in research preview on April 14, 2026. A few things to be aware of:
- GitHub webhooks have per-routine and per-account hourly caps — events beyond the limit are dropped until the window resets
- Routines belong to your individual account — they are not shared with teammates, and actions appear under your GitHub and connector identity
- Routines are not persistent AI agents — each run is a fresh session with no memory of previous runs. If you need state across runs, you'll need to store it in the repo or an external system
- API surface may change — Anthropic has stated that behavior, limits, and the API surface may change before general availability
- Webhook support beyond GitHub is planned but not yet available
Frequently Asked Questions
Do Claude Routines work with all programming languages?
Yes. Routines are language-agnostic — they're instruction sets for Claude, not code themselves. You can use them for Python, TypeScript, Rust, Go, or any language Claude Code supports.
Can I use Routines on an existing project?
Yes, and that's the most common use case. You add the repository to the routine and Claude clones it fresh at the start of each run.
What happens if Claude Routines pushes bad code?
Routines create claude/-prefixed branches for changes — they don't push directly to main unless you explicitly enable unrestricted branch pushes. You review and merge (or discard) like any other PR.
Does /schedule in the CLI still work?
Yes, but its behavior changed: /schedule now creates cloud Routines with schedule-type triggers (not session-scoped tasks). The old session-scoped /loop behavior is still available via /loop.
Can multiple triggers fire the same routine?
Yes. A single routine can have a schedule trigger, an API trigger, and a GitHub event trigger active simultaneously — the same configuration runs regardless of which trigger fires.
Getting Started Today
Claude Routines are available now for all paid Claude Code subscribers. The fastest path to your first routine:
- Go to claude.ai/code/routines
- Click New routine
- Start with a simple use case — nightly issue labeling or a PR review trigger
- Use Run now to test before setting a live trigger
The feature is in research preview, so behavior may evolve — but the core pattern is already stable enough to build real workflows on.
Published April 16, 2026 · Updated as new information becomes available
Tags: Claude Code, Claude Routines, AI Automation, Developer Workflow, Anthropic
