How to Use Claude Code Routines | 3 Trigger Examples and Pricing

What Is Claude Code Routines — 3 Trigger Types and Its Role as "Claude Code Running in the Cloud"

Claude Code Routines is an add-on feature for Pro and Max plans that runs Claude Code in the cloud. Unlike the Claude Code you interact with in a local terminal, the defining characteristic is that processing continues even when your PC is closed. There are three trigger types:

• Schedule trigger: Fires at a specified time
• API trigger: Fires when called via HTTP from an external source
• GitHub event trigger: Fires in response to events like PR opens or pushes

The differences between these three are covered in detail with real-world scenarios later in this article under "Examples by Trigger Type." Output results can be viewed chronologically in the Routines tab of the web or desktop app (source).

Not the Same as In-Session Scheduling — Clearing Up a Common Confusion

A feature that's easy to confuse with Routines is in-session scheduling, where Claude Code itself proposes a time to resume processing during an active conversation. This works by Claude Code suggesting "please resume at ◯◯:00" and ending the dialogue — it does not have the property of "continuing to run even when your PC is closed" that Routines offers (source). The key thing to understand upfront is that Routines is a separate layer that lives persistently in the cloud.

Getting Started — Enabling Routines from the Web or Desktop App and the Fastest Path to Your First Running Routine in 10 Minutes

The standard approach for your first Routine is to get it to a "working" state as quickly as possible, then refine the content afterward. To get a single Routine running in 10 minutes, follow these five steps. The steps themselves are simple, but two decisions — your initial trigger choice and the granularity of your prompt — are worth thinking through carefully, as they determine how smoothly operations go afterward. Improving your first Routine after it's running is worth 10 times the effort of getting the initial setup right (source).

5 Steps to Get Your First Routine Running in 10 Minutes

1. Open the Routines tab in the web or desktop app.
2. Create a new routine and give it a name you'll recognize later (e.g., daily-doc-diff — names that convey the content and frequency are easier to manage).
3. Select a trigger: Schedule, API, or GitHub. For your first Routine, choosing "Schedule trigger, once a day" minimizes the impact of any failures and makes testing easier (source).
4. Write the prompt for Claude Code. To start, simply specifying three things is enough: the path to the target file, what you want it to read, and the output format.
5. Run a test execution to verify it works. If it fails, open the history tab and check the logs to see where it stopped (source). Making it a habit to run at least one test before going live will save you from surprises in production.

Examples by Trigger Type — When to Use Schedule / API / GitHub Event Triggers for Real Work Scenarios

The three triggers differ in when they fire. Schedule triggers fire at a time, API triggers fire on external events, and GitHub event triggers fire on repository events. In practice, the right approach isn't "pick a trigger that fits a work scenario" but rather "clarify the firing conditions for the task you want to put in Routines, then pick the one trigger that matches." For each scenario below, we include one line on why Routines is a good fit and one line on where it falls short (source).

Schedule Trigger: "Daily Doc Diff Summary" Every Morning at 7 AM

In teams where public-facing product documentation is updated daily, having Routines generate a summary of the previous day's changes every morning at 7 AM is a useful practice. Readers can open the Routines tab right when they arrive at work and get up to speed on the day's release notes in 30 seconds (source). Why it's a good fit: the task is "summarize a fixed subject at a fixed time." Why it falls short: it can't handle tasks that require human judgment partway through — those should stay in the local interactive session.

API Trigger: Drafting Initial Replies to Tagged Messages in Slack or a Shared Inbox

Teams that manage inquiries through Slack or shared email can speed up initial response times by calling Routines via API the moment a specific tag or keyword appears, generating an "inquiry summary and draft initial reply." Why it's a good fit: the trigger is driven by external events. Why it falls short: if you try to have the Routine complete the process all the way to sending the response, the accountability burden when things go wrong becomes heavy — the safer design is to have a human review the Routines output before it's sent (source).

GitHub Event Trigger: "Lightweight Change Summary and Risk Notes" Posted Ahead of PR Review

The classic use of GitHub event triggers is firing Routines the moment a PR is opened or a push occurs, leaving a summary of the changes and a note on potential concerns as a PR comment. It works well as prep work to orient reviewers before a full code review. Why it's a good fit: it's event-driven and finishes quickly. Why it falls short: if you try to replace a full code review with it, things get missed — keep Routines to summarizing, organizing, and tagging (source).

Pro: 5 Runs/Day / Max: 15 Runs/Day — Execution Limits, Real Pay-as-You-Go Costs, and Designing Operations That Don't Blow Your Budget

Routines execution is limited to 5 runs per day on Pro and 15 per day on Max within the free tier. Beyond that, pay-as-you-go charges apply — so the most cost-efficient design is to prioritize lightweight, regularly scheduled tasks for Routines and keep ad-hoc heavy processing in your local interactive session (source). Note that whether these limits apply to your total daily runs across all routines or per individual routine, and whether test runs or failed runs count toward the limit, are important assumptions when designing your usage management. Since the counting unit and how failures and test runs are handled can vary by plan and change over time, always refer to both the official Anthropic documentation and third-party real-world usage reports as primary sources when looking at specific pricing figures.

How to Allocate "5 Runs" or "15 Runs" — Think in Three Blocks: Morning, Midday, Evening

With Pro's 5 runs, a split of "2 morning / 1 midday / 2 evening" works well; with Max's 15, "5 morning / 5 midday / 5 evening" lets you either use up the full limit or keep 1–2 slots in reserve for unexpected tasks. If you're frequently hitting the limit on Pro, the next decision is either upgrading to Max or offloading some tasks to OS-level scheduling (source).

Opus 4.7 Fast Mode Now the Default — Factor in Fast Mode Costs for Heavy Processing

As of the May 12, 2026 release, Opus 4.7 Fast Mode is available as the default model (source). If you're routing heavy processing through Routines, make sure to check the Fast Mode rates and actual costs as well, and estimate a monthly budget that accounts for the free tier plus potential pay-as-you-go charges. Setting a budget cap at the start of each month and reviewing variance at month's end creates a cycle where you can make informed decisions — on days when a heavy ad-hoc task risks exceeding your budget, you can fall back to local interactive mode. Note that default models and Fast Mode availability are updated frequently, so this section reflects a snapshot as of May 2026; always check the official release notes for the latest default model.

Choosing Between Routines and Alternatives — How to Decide Between Routines, Always-On Scripts, and Third-Party Schedulers (Frequency × Weight × Failure Cost)

Routines excels at running lightweight tasks in the cloud, but it's not the right answer for every scheduled process. The real work of operations design is narrowing down what belongs in Routines after comparing it against OS-level scheduling, managed persistent processes, and third-party schedulers. Evaluate on three axes: frequency, weight, and failure cost. Moving tasks into Routines in order from those that satisfy all three criteria makes for a manageable migration (source).

Axis 1: Frequency — Does It Fit Within the 5–15 Runs/Day Limit?

Tasks that fit within 5–15 runs per day are prime candidates for Routines. Anything beyond that (e.g., hourly monitoring, health checks every 5 minutes) is better distributed across OS-level scheduling or managed persistent processes — this keeps your Routines free tier focused on "output worth a human reading" (source).

Axis 2: Weight — Long Single-Run Processes Are a Poor Fit for Routines

Single-run heavy processes tend to conflict with Routines' limits, timeouts, and cost model. The cleanest division: put tasks that need only brief instructions and produce summaries, diffs, or formatted output into Routines; leave heavy processing on local machines or managed environments. This keeps operational costs at their lowest (source).

Axis 3: Failure Cost — Don't Use Routines as the Sole Actor for Tasks Requiring Immediate Human Intervention

Tasks where failure demands immediate human intervention — production incident response, final dispatch of customer notifications, payment processing — should never be completed entirely by Routines acting alone. Keep Routines scoped to notifications, summaries, prep work, and draft generation; final decisions stay with a human. Documenting recovery procedures in advance is also an important operational safeguard (source).

Checklist: Deciding on Your "First Routine" to Run Today

• Pick one lightweight recurring task you do at the same time every day (target: under 5 minutes / acceptable if it fails).
• Rewrite that task as a Routines schedule trigger and run a test execution.
• After one week of operation, check how much time you recovered and what the actual pay-as-you-go charges came to.

Once your first Routine is running, everything after that is just repeating the same pattern. The next step in this article starts with opening your own work calendar and picking one task.

Sources

• Routines — Official Anthropic Documentation (Japanese)
• Platform Release Notes (May 12 — Fast Mode Opus 4.7)
• Anthropic Official News — Claude for Small Business
• Qiita: Three Days Running Claude Code Routines
• Zenn: Claude Code Routines Schedule Trigger Implementation Example
• uravation.com: Claude Code Routines Desktop UI Walkthrough