ccr-cloud documentation
ccr-cloud lets you run Claude Code tasks on your own server and control them from a browser or phone. Start a task, close your laptop, and come back to its progress, approvals and results later. Your code, model access and Git credentials stay on your server.
New here? Start with Getting started. It takes about five minutes on a server where Claude Code already works.
#Get started
- Getting started: install, connect from your laptop and run your first task.
- How it works: the control plane, runners, execution modes and what each one can access.
#Guides
- Work with repositories: use checkouts already on the server, or add a repository by URL.
- Run and control tasks: permissions, response style, approvals, follow-ups, stop and resume.
- Switch models with CCR: use another model, such as an open-source one, while you're at a Claude usage limit.
- Access from your phone: Tailscale, your own domain with Cloudflare, or a VPN you already run.
- Notifications: approval and completion alerts on your devices.
- Discord control: an optional owner-only Discord bot.
- Docker and CCR sessions: isolated containers through a Claude Code Router gateway.
- Operate your server: updates, restarts, backups, logs and host resources.
#Reference
- Configuration: every setting in
server.env. - Command line:
setup,doctor,github,resourcesanddev. - HTTP API: routes for scripts and other clients.
- Troubleshooting: symptoms, checks and fixes.
#Status and proposals
ccr-cloud is a single-owner prototype. Everything in the guides above is available today. Tasks run Claude Code; Codex isn't supported as a task agent yet (#19). These documents describe work that is proposed and not implemented:
- Optional conversational dispatcher
- Managed connection and its roadmap
- Multi-agent Rooms (RFC #20)
Older links to the user guide and the operations reference still work; those pages point to where each section moved.