Claude Code Routines and Mono Agent: how they fit together
Claude Code Routines run Claude Code sessions in Anthropic's cloud on a cron schedule (at most hourly), an API call or a GitHub event, and Claude in Chrome adds scheduled shortcuts and recorded browser workflows. Every run is driven by the model. Mono Agent complements them: deterministic steps, schedules finer than an hour, webhooks, durable human approval, local run history and recorded browser steps that replay without a model call, with AI only where it adds value via agent.ask on Claude Code itself.
| Claude Code Routines | Claude in Chrome | Mono Agent | |
|---|---|---|---|
| Where it runs | Anthropic cloud | Your browser | Your machine |
| Triggers | Cron (minimum 1 hour), API call, GitHub events | Scheduled shortcuts | Manual, cron schedule (down to seconds), webhook, agent org |
| Each step | Model-driven | Model-driven | Deterministic, with agent.ask where you want AI |
| Human approval | Not a built-in queue | Not a built-in queue | core.human_in_loop: durable queue, edit before approve, timeout |
| Run history | In Claude Code | In the browser | Local SQLite |
| Status | Research preview since 2026-04-14 | Generally available for paid plans since 2026-08-26 | Released; monoes.me sign-in required |
What Routines and Claude in Chrome do well
A Routine is a saved Claude Code task that runs on Anthropic's infrastructure on a schedule, from an API call or on a GitHub event, so nothing on your side needs to stay on (overview: https://www.dsebastien.net/claude-code-routines/). Claude in Chrome brings scheduled shortcuts and workflow recording to the browser you use every day. Both treat the model as the engine: every run is an agent session that reads the situation and decides what to do.
What Mono Agent adds
- Deterministic steps: HTTP calls, data transforms, branching and database writes run the same way every time and cost no model tokens.
- Schedules finer than an hour: the scheduler accepts cron expressions with seconds.
- Webhooks: any service that can send an HTTP request can start a workflow.
- Durable human approval: core.human_in_loop holds a run until a person approves, edits or rejects it, and the queue survives restarts.
- Local run history: every run and its per-step output is stored in SQLite on your machine.
- Recorded browser steps: the recorder turns your clicks in your own logged-in Chrome or Edge into an automation that replays without a model call per run; browser.jev adds an adaptive step only where a page changes.
- AI where it adds value: agent.ask runs a prompt through Claude Code itself, on your existing plan, for the steps that need judgment.
How to connect them
Register Mono Agent's MCP server with Claude Code: claude mcp add monoagent -- monoagentcli mcp --allow-mutations. Claude Code can then list workflows, run them and approve or reject items waiting in the human-approval queue. Without --allow-mutations the server is read-only: Claude Code can list and inspect, but not run or approve. monoagentcli setup can also register the server for you; check with claude mcp list which flags it was registered with.
This works for Claude Code running on the same machine as Mono Agent. A Routine running in Anthropic's cloud cannot reach a local stdio MCP server; to start a Mono Agent workflow from a cloud Routine, have it call a webhook you expose. In the other direction, a Mono Agent workflow can start a Routine through its API trigger with an HTTP step.
When Routines alone is simpler
- The task is mostly judgment: triaging issues, reviewing a pull request, summarizing a repo. A Routine does that with a prompt and no workflow to build.
- Hourly or daily is often enough and you do not want a machine of your own running.
- The trigger is a GitHub event; Routines handle that natively.
- Nothing needs a human sign-off before it happens.
- You do not want another account: Mono Agent needs a monoes.me sign-in.
Frequently asked questions
Does Mono Agent replace Claude Code Routines?
No. Routines are good at model-driven tasks in the cloud. Mono Agent handles the deterministic parts, approvals and frequent schedules, and can call Claude Code through agent.ask where a step needs AI.
Can a Routine run a Mono Agent workflow?
Through a webhook you expose, yes. The MCP route (monoagentcli mcp) works only for Claude Code running on the same machine as Mono Agent, because Routines run in Anthropic's cloud.
Does agent.ask use my Claude subscription?
It runs Claude Code with Claude Code's own login, so it uses whatever plan that login has, including its usage limits. Mono Agent adds no AI charge.
How often can a Mono Agent workflow run?
Cron schedules accept seconds, so well under an hour, as long as the daemon is running. Routines' minimum is one hour.