<p>Anthropic has launched Routines in Claude Code, currently in research preview for Pro, Max, Team, and Enterprise subscribers. The short version: you configure a task once — a prompt, a repo, and any connectors — and it runs automatically on Anthropic’s cloud infrastructure, no local machine required.</p>
<h2>What routines actually are</h2>
<p>Routines are saved Claude Code configurations that run as full cloud sessions. They replace the older <code>/schedule</code> command (existing scheduled tasks migrate automatically) and extend it with two additional trigger types:</p>
<ul>
<li><strong>Scheduled</strong> — hourly, daily, or weekly</li>
<li><strong>API</strong> — HTTP POST to a per-routine endpoint with bearer token auth</li>
<li><strong>GitHub events</strong> — pull requests, pushes, releases, and issues</li>
</ul>
<p>You can combine multiple trigger types on a single routine. When a routine runs, it has access to connected MCP servers, GitHub repositories, and configured environment variables. There are no approval prompts — it runs fully autonomously — so prompt quality and scope matter a lot. Branch writes default to a <code>claude/</code> prefix to avoid clobbering long-lived branches.</p>
<h2>What you can use them for</h2>
<p>Anthropic’s documented use cases include:</p>
<ul>
<li>Nightly backlog triage and issue labeling</li>
<li>Automated code review on pull requests</li>
<li>Deploy verification and CI/CD smoke checks</li>
<li>Documentation drift detection</li>
<li>Alert triage</li>
<li>Cross-SDK library porting</li>
</ul>
<p>The per-PR continuity is worth noting: a routine can trigger on each new pull request and maintain context across that PR’s lifecycle, which is different from a one-shot cron job.</p>
<h2>Plan limits</h2>
<p>Routines count against a daily run allowance that varies by plan:</p>
<table>
<thead>
<tr>
<th>Plan</th>
<th>Daily routine limit</th>
<th>Monthly cost</th>
</tr>
</thead>
<tbody>
<tr>
<td>Pro</td>
<td>5</td>
<td>$20</td>
</tr>
<tr>
<td>Max</td>
<td>15–25</td>
<td>$100–$200</td>
</tr>
<tr>
<td>Team/Enterprise</td>
<td>25</td>
<td>varies</td>
</tr>
</tbody>
</table>
<p>Routines run against your subscription rather than API billing, which makes them cheaper for recurring tasks than calling the API directly.</p>
<h2>How to set one up</h2>
<p>You can create routines from <a href="http://claude.ai/code" target="_blank" rel="noopener nofollow">claude.ai/code</a>, the CLI (<code>/schedule</code>), or the desktop app. The setup asks for a name, a system prompt, a schedule or trigger, and any environment variables. The older <code>/loop</code> command still works for local, session-scoped recurring tasks if you want something that stays on your machine.</p>
<h2>Caveats worth knowing</h2>
<p>A few things to keep in mind before going all-in:</p>
<p><strong>Latency.</strong> LLM inference takes 2–15 seconds per step. Routines aren’t suitable for real-time webhooks or anything that needs a sub-second response.</p>
<p><strong>No pre-built OAuth connectors.</strong> Unlike <a href="http://Make.com" target="_blank" rel="noopener nofollow">Make.com</a> or n8n, there’s no library of pre-wired integrations. You’re working with MCP servers and whatever your prompt can reach.</p>
<p><strong>Prompt discipline matters more here.</strong> Because routines run unattended and autonomously, a vague or overly broad prompt can cause real damage. Scope your prompts carefully.</p>
<p><strong>Some developers prefer GitHub Actions.</strong> One engineer noted that GHA gives more control over secrets management and execution environment — and that Claude is good enough at writing the YAML that the setup cost is low anyway. Routines are convenient, but they’re not the only option.</p>
<h2>The bigger picture</h2>
<p>Routines are Anthropic’s move into the workflow automation space that tools like <a href="http://Make.com" target="_blank" rel="noopener nofollow">Make.com</a> and n8n currently occupy. The developer-native angle — plain prompts, Git integration, no proprietary node editor — is genuinely different from those tools. Whether it’s better depends on what you need: if you want pre-built connectors and a visual editor, existing tools still win. If you want AI reasoning baked into the automation loop and you’re already living in Claude Code, routines are worth trying.</p>
<p>The research preview label means the feature is still evolving. Existing <code>/schedule</code> tasks have already been migrated automatically.</p>


Collections

Anthropic has launched Routines in Claude Code, a cloud-based automation feature currently in research preview for Pro, Max, Team, and Enterprise subscribers. Routines let you configure a task once — with a prompt, repo, and connectors — and have it run automatically on Anthropic's infrastructure without a local machine. Three trigger types are supported: scheduled (hourly/daily/weekly), API (HTTP POST with bearer token), and GitHub events (PRs, pushes, releases, issues). Multiple triggers can be combined on a single routine. Routines run fully autonomously with access to MCP servers, GitHub repos, and environment variables. Daily run limits vary by plan (5 for Pro, 15–25 for Max, 25 for Team/Enterprise) and run against subscription billing rather than API billing. Key caveats include LLM inference latency (2–15s per step), no pre-built OAuth connectors, and the need for careful prompt scoping since routines run unattended. Anthropic positions this as a developer-native alternative to tools like Make.com and n8n, with plain prompts and Git integration instead of a visual node editor.

Claude Code now supports automated routines triggered by schedule, API, or GitHub events