Corral

Let Claude Code work on your project on autopilot — without worrying about what it might touch. Corral runs Claude with permission prompts turned off, but inside a safe bubble: it can only reach the sites you allow, it never sees your real credentials, and everything happens in a throwaway container that leaves your machine untouched. You watch it work in a live dashboard in your browser.

Try it

# 1. install (Docker required; the installer adds its other host deps for you)
curl -fsSL https://raw.githubusercontent.com/scoutapp/corral/main/scripts/install.sh | bash

# 2. two commands get you going
corral populate-proxy-credentials   # set your credentials (once)
corral start                        # start Claude + open the dashboard

start prints a private link to the browser dashboard. From there, hit New project to spin up a sandbox on a repo (or a blank/existing directory) and start working — you watch Claude live, see what it's reaching out to, and drop into a terminal, with one page covering all your projects at once.

Prefer the terminal? You can also drive a single project from the CLI: cd into it, run corral init once to set up ./.corral/, then corral start runs Claude right there in that directory.

→ New here? Read the usage guide — a screenshot tour of the dashboard and everything you can drive from your browser. It's the best place to start.

What keeps it safe

• Nothing leaks. Claude runs with a dummy token; your real credentials are swapped in behind the scenes and never enter its environment.
• Nothing unexpected gets out. Outbound network is limited to the sites you allow — including anything Claude spins up in Docker. (start is permissive at first so a new project just works, quietly recording what it uses; you lock it down to that list when ready — see the usage guide.)
• Nothing sticks around. It all runs in a disposable container; close it and your machine is as it was.
• You're always watching. The live dashboard shows exactly what's happening.

For the full trust model — including the deliberate trade-offs and residual risks (the privileged DinD container, loopback reachability, dangerous mode) — see docs/security.md.

Commands

The ones you'll use most:

Command	What it does
populate-proxy-credentials	set your credentials (once)
start	start Claude + open the dashboard
dashboard	open the dashboard on its own
init	set up ./.corral/ in a directory (optional — only for the per-project CLI flow)
update	update Corral itself
uninstall	remove Corral from your machine
help	full command list and options

For the complete list, see the usage guide or run corral help.

Good to know

• You need Docker and Claude Code signed in (claude once). The installer pulls in its other host deps for you — mitmproxy (the credential proxy) and tmux (hosts the session) — via Homebrew or apt.
• Everything for a project lives in ./.corral/ (safe to delete to start over; init git-ignores it for you).
• Shared settings and credentials live in ~/.corral/.
• Pin a version or change install locations with CORRAL_VERSION, CORRAL_PREFIX, CORRAL_HOME before the install command.

For developers

Curious how it works under the hood, or want to hack on Corral itself? Start with docs/developers.md — building from source, running the tests, the release process, and how to actually iterate on Corral. The design is in docs/architecture.md.

git clone https://github.com/scoutapp/corral.git && cd corral
./install.sh                     # build + install from source
go build -o corral ./cmd/corral && ./corral list   # or run from the checkout
go test ./... && (cd tests/e2e && npm test)                    # tests (e2e also runs in CI)

License

MIT © Scout Monitoring