SCC CLI

Start Here

Most readers do not need to browse the whole docs tree.

If you just want to get productive quickly, follow this path:

1. Quick Start Install SCC, connect Claude and/or Codex, and launch your first sandboxed session

2. Core Concepts Understand providers, sessions, profiles, worktrees, and network policy

3. Daily Workflow Learn the commands you will actually use every day: start, resume, stop, list, and worktrees

If you are evaluating SCC rather than using it day to day, start with What is SCC?, then read Security Model and Architecture Overview.

Find Your Path

I'm joining a team Get running in 30 seconds with your team's config

I manage a team Configure plugins and profiles for your developers

I'm an org admin Set up security policies and governance

I'm evaluating SCC Learn what SCC does and why

Popular Topics

These pages answer the questions that usually bring new readers to SCC in the first place.

Why sandbox AI coding agents? Understand the real risks, where containers help, and where network controls matter

AI coding agent governance See how organizations roll out Claude, Codex, or both without losing control

Network controls for AI coding agents Understand why container isolation is not enough on its own

How SCC handles network access See how open, web-egress-enforced, and locked-down-web differ in practice

Run agents in containers Understand the container model, first-run image builds, and runtime choices

Compare SCC with local agent use See the tradeoffs between local convenience and governed team rollout

Why SCC?

Multi-Provider Runtime

Teams can standardize on Claude, Codex, or both. Developers can choose the agent that fits the task while keeping the same governance, profiles, and audit surface.

Container Isolation

Code runs in OCI container sandboxes. The agent only sees the mounted workspace instead of your full host filesystem.

Network Egress Control

Containers alone do not stop an agent from reaching the network. SCC can leave egress open, force HTTP/HTTPS through a proxy sidecar, or disable network access entirely.

Standardized Configs

Organization-managed profiles distributed via a single URL. No manual setup for developers.

Built-in Safety Engine

Fail-closed wrappers block destructive git commands and intercept network tools inside every container, even if an agent-native plugin is missing or disabled.

If the main concern is what an agent can reach on your network, start with Security Model or Why Sandbox AI Coding Agents. Containers help, but they are not the whole boundary.

Quick Start

1. Install SCC CLI

   uv (recommended)

   uv tool install scc-cli

   pipx

   pipx install scc-cli

   pip

   pip install scc-cli

2. Verify your environment

   scc doctor

3. Configure and launch

   scc setup    # Paste your org config URL, connect providers
   scc start    # Launch in current directory

SCC setup can connect Claude, Codex, or both. If your organization allows both, developers can keep a default or ask each time and choose the right provider per task.

Who Controls What

Setting	Org Admin	Team Lead	Developer
Block dangerous plugins	✅ Sets	❌ Cannot override	❌ Cannot override
Default plugins for all teams	✅ Sets	—	—
Team-specific plugins	✅ Approves	✅ Chooses	—
Project-local config (.scc.yaml)	✅ Can restrict	✅ Can restrict	✅ Extends
Safety-net policy	✅ Sets	❌ Cannot override	❌ Cannot override
Provider choice	✅ Sets allowed list	—	✅ Chooses from allowed

Learn more about governance Understand the full delegation and security model