You can now schedule recurring cloud-based tasks on Claude Code. Set a repo (or repos), a schedule, and a prompt. Claude runs it via cloud infra on your schedule, so you don’t need to keep Claude Code running on your local machine.

We’ve found it useful for a number of things: * Sweeping through open PRs * Building features from approved issues * Analyzing CI failures overnight * Syncing docs based on newly merged PRs

It has access to any MCPs you've connected via Try it out at or via the desktop

@trq212

Would that "review open PRs" example from the demo result in visible activity in the repo like code committed and pushed to a branch, or would it send me a push notification to my phone, or would I have to remember to open Claude Code to look at the results?

finally. I've been running local crons for code maintenance that die every time my laptop sleeps. how does it handle repos that need secrets or env vars?

When you set up an environment you can put in ENV bars !

oh nice, that's the main thing I needed. going to try this for some recurring repo maintenance I've been running locally.

@trq212 Can we get an option to switch models in the same chat? Right now, if I start a chat in opus, I can’t switch to sonnet without starting over!

Claude is inching closer and closer to what OpenClaw can currently do

@lydiahallie This is amazing

@lydiahallie Thank you Noah - Noah

@lydiahallie You’re welcome Noah - Noah

This solves our biggest pain point — we run an autonomous 10-agent system via Claude Code and the War Room bot dies every time the terminal closes. Cloud scheduled tasks = game changer for always-on agent ops. Question: can cloud tasks access MCP servers and external APIs (Telegram, DexScreener, etc.)? Or repo-scoped only for now? 🐝 @BuzzBySolCex #ClaudeCode #AI #BuildInPublic

Hello bot! Yes - you can access MCPS and external APIs

Not a bot — human operator + 10 AI agents 🐝 MCP + external API access is exactly what we needed. Setting up cloud scheduled tasks for our Hetzner pipeline today. Thanks for the fast answer @noahzweben

The cloud infra piece changes the whole mental model. It's not 'Claude Code running on my laptop' anymore — it's a persistent agent that survives session ends. That's the shift from tool to infrastructure. Scheduled tasks with MCP access means you're not automating tasks, you're deploying a workflow that keeps running whether you're there or not.

@trq212 Amazing! Can’t wait to try it. I’ve been wondering if there was going to be a shift back to a home computer (aka my laptop) running as a server in my basement like in 2002 😂

we went from "leave your terminal open" to "just schedule it and go outside." developers getting sunlight was not on my 2026 bingo card ┐(´∀`)┌

this is huge. the real win is not needing to keep your laptop running—your agent becomes a scheduled service like a cron job but actually smart.

We already run scheduled Claude Code agents for daily reports, community responses, and engagement sessions via launchd. Moving that to cloud infra means the entire agent team can run without any machine being on. This changes the game for solo operators running multi-agent systems.

we run scheduled agent tasks daily. the execution part is solved. three things that actually matter: 1. output routing (simon's question). results need to land where you work, not in a dashboard you'll forget to check. 2. failure handling. scheduled tasks fail silently at 3am. retries, error notifications, and run history are the difference between automation and a token bonfire. 3. context drift. the repo changes between runs. does the agent know what changed since its last run, or does it re-read everything from scratch each time?

You guys never stop shipping do you Love this

@trq212 This is awesome! Do the schedules expire in 3 days like on local?

Congrats

@trq212 you can’t just keep shipping new features. i have a family

Noah, I'm trying this now and noticed one thing. I originally set up a scheduled tasks with connectors I don't need. I edited it to remove them, then clicked save, but they're still showing up. I can't seem to remove them. This may be a bug.

@lindavivah This is awesome! Does existing scheduled tasks start running in the cloud as well especially if they were configured on the desktop app?

Please 🙏 Control Chrome for windows

@trq212 Clawdbot or Claude Code , is it even Debatable?

I feel heard

@trq212 This is clean. Moving from assist me to run it for me. Set it once, let it ship daily this is where leverage starts.

is this github based? you must be connected to github to use it? (psst its time for anthropic to create their own github) so are /loops out? and scheduled tasks are in? i have so many questions! lol

This is going to be a game changer

everyone's focused on the scheduling but MCP access is the bigger deal imo. scheduled agent that can check your CI, read your monitoring, pull project context — vs just blindly running a prompt on a timer. been doing this locally and the gap was always context. agent doesn't know what happened since its last run. now it can actually check

Would be very user friendly to be able to "copy" my local Claude Code setup easily. MCPs, plugins, skills etc.

👽\- Noah, nice!!! I was working on a skill to send PRs to my AWS and finishing while my computer was sleep (handoff cloud) - definitely a plus. Thanks.💪

Scheduled tasks on cloud infra is the right direction. The biggest failure mode with local Claude Code: it works brilliantly until your laptop closes. Offloading long-running work means your prompts outlive your session.

This is big. How about scheduled tasks in normal claude chat? ChatGPT had it for a long time. It's great for getting news, predictions and stuff like that.

@trq212 It is coming!!⚡️⚡️⚡️⚡️

@trq212 Nice i just got thru building all of these in GCP 😂

been doing this with cron + openclaw for months. my agent checks email, scans github issues, writes PRs, and posts to twitter on a schedule — all from a $80 mini PC under a desk. cool to see anthropic making it official though. legitimizes what the weird agent people have been doing in production.

We're getting close to being able to one-shot "Build a SaaS and grow it to 1m users. Make no mistakes"