Claude Code routines are landing in research preview as a new automation layer designed to run Claude-driven workflows without the usual background babysitting. The idea is straightforward: define an automation once—prompt, repo, and connectors—then run it on a schedule, trigger it via an API call, or fire it from an event. Because routines run on Claude Code’s web infrastructure, they don’t depend on a local machine staying online.
In practice, routines package up work that previously meant owning the glue: cron, infrastructure, and extra tooling such as MCP servers. With routines, that packaging also includes access to repositories and connectors, so the automation can act with the same context Claude Code already uses interactively.
Three ways to trigger a routine
Claude Code routines come in three flavors, each aimed at a different “automation entry point” teams tend to build around.
Scheduled routines
Scheduled routines run on a cadence—hourly, nightly, weekly—based on a prompt. Anthropic notes that tasks previously scheduled via /schedule in the CLI are now considered scheduled routines.
A representative example: a nightly job that pulls the top bug from Linear, attempts a fix, and opens a draft PR.
API routines
API routines are designed for systems that can make an HTTP request. Each routine gets its own endpoint and auth token; a POST returns a session URL, which makes it possible to pipe in payloads from deploy hooks, alerting systems, or internal tools.
One example prompt focuses on operational triage: ingesting an alert payload, identifying the owning service, then posting a summary and proposed first step to an on-call channel.
Webhook routines (starting with GitHub)
Webhook routines can subscribe to GitHub repository events and run automatically based on filters, creating a new session per matching PR. Claude Code also continues to feed PR updates—like comments and CI failures—into that session for follow-up handling. Anthropic says webhook-based routines will expand to additional event sources over time.
Common patterns emerging in early usage
Anthropic points to a few routine “shapes” that early users are building:
- Backlog management and docs drift checks as scheduled routines
- Deploy verification, alert triage, and feedback resolution via API-triggered routines
- GitHub-triggered workflows like cross-language library ports and team-specific code review checklists, with inline comments produced before a human reviewer steps in
Across these examples, the through-line is tooling that stays close to the repo while also reaching outward through connectors and event triggers.
Availability and limits
Routines are available to Claude Code users on Pro, Max, Team, and Enterprise plans, as long as Claude Code on the web is enabled. Creation happens at claude.ai/code, and scheduling can also be initiated via /schedule in the CLI.
Usage follows subscription limits like interactive sessions, with additional daily routine caps:
- Pro: up to 5 routines/day
- Max: up to 15 routines/day
- Team/Enterprise: up to 25 routines/day
Additional runs beyond the daily limits are available with extra usage, with more details in the routines docs.
