Claude Code Remote Session Manager (Multi-Provider)

Run Claude Code with Remote Control on your own VPS, fully Dockerized, with Claude, DeepSeek, or any Anthropic-compatible model.

Sign in with GitHub, register an Account once, and spin up as many sandboxed agent Sessions as you like from the browser: each one its own container, its own cloned workspace, and its own built-in web terminal. Nothing is installed on the host but Docker; no agent container ever touches a host path.

Not just Claude. An Account can be your Claude subscription (OAuth login, with Remote Control), DeepSeek, or any Anthropic-compatible endpoint: you supply the API key, base URL and model. Run different Sessions on different providers side by side. Remote Control is a Claude feature; with the others you drive the agent from the web terminal (that is a DeepSeek session in the screenshot above).

Warning

The main branch is under active development and may be unstable. Deploy the latest tagged release (currently v1.0.0-alpha.1), not main.

Features

• Multi-Session web manager: create, stop, reset and destroy Claude Code containers from the browser, each with a built-in web terminal.
• Accounts, not host credentials: authenticate each Account once (interactive claude OAuth login in the browser, or an API key); credentials live in a Docker volume, never on the host.
• Bring your own provider: Claude, DeepSeek, or any Anthropic-compatible endpoint (your key, base URL and model). Sessions on different providers run side by side; Remote Control is Claude-only, the web terminal works with all of them.
• Sandboxed workspaces: every Session clones its GitHub repo into an isolated Docker volume.
• GitHub sign-in with a fail-closed allow-list: only the usernames you list get in; an empty list denies everyone.
• Automatic HTTPS: optional Caddy reverse proxy with Let's Encrypt.
• Auto Mode with real isolation: agents run --permission-mode auto inside locked-down containers with no host mounts, a split Docker network, and memory/CPU/PID caps. Read Security & sandboxing before trusting it with anything.

Quick start

You need a VPS with Docker, Docker Compose and Git, plus a GitHub App (takes a few minutes; the install guide walks you through it). No Claude Code install on the host.

git clone --branch v1.0.0-alpha.1 https://github.com/sgomez/cc-remote.git
cd cc-remote
./setup.sh

The wizard asks for your domain, the GitHub App details and the allowed GitHub usernames, derives everything else, and offers to start the stack. If you skipped that last step:

docker compose up -d --build

Open https://<your-domain> (or http://localhost:4000 locally), sign in with GitHub, register an Account, and create your first Session.

First time? Follow the step-by-step install guide.

Updating

Your settings (config.json, .env) are gitignored and survive updates, and running Sessions are left untouched:

git fetch --tags && git checkout v1.0.0-alpha.1
docker compose up -d --build

Coming from v0.1.0

Don't migrate. Start clean. Push any work you care about to GitHub first (this deletes every workspace), then wipe everything cc-remote from Docker and set up again:

docker compose down -v
docker ps -aq --filter "name=cc-remote" | xargs -r docker rm -f
docker volume ls -q --filter "name=cc-remote" | xargs -r docker volume rm
docker network rm cc-remote 2>/dev/null || true
git fetch --tags && git checkout v1.0.0-alpha.1
./setup.sh

Documentation

• Install guide: every step spelled out, from the GitHub App to your first Session.
• User guide: Accounts, Sessions, the web terminal, day-to-day commands, and configuration.
• Security & sandboxing: what makes Auto Mode acceptable here, and what it does not protect you from.