CC Switch: how to switch Claude Code and Codex between API providers
What CC Switch is, how to install it from the official repo, how to use it with Claude Code and Codex, and what to watch out for — data, terms and lost features.
Claude Code talks to Anthropic, Codex to OpenAI — unless you edit their config files by hand. CC Switch is a free, open-source desktop app that does that editing for you: pick a provider, paste a key, click, and your coding agent now runs on a different model. Here is what CC Switch is, how to install it safely, how to use it with Claude Code and Codex, and what to watch out for.
1. What is CC Switch
CC Switch is a cross-platform desktop app (Windows, macOS, Linux) for
managing and switching API providers across AI coding tools. Instead of
editing JSON, TOML or .env files, you keep a list of providers in the app
and switch between them with one click — from the main window or the system tray.
It supports Claude Code, Claude Desktop, Codex, Gemini CLI, OpenCode and several more agents, ships with dozens of provider presets, and is MIT-licensed. Switching replaces only the connection details — endpoint, key and model — and leaves the rest of your configuration alone.
2. Why developers use it
- Cost. Run routine work on a cheaper model and keep the expensive one for hard problems.
- Trying models. Compare how the same agent behaves on different models without touching config files.
- Availability. Fall back to another provider when one is down or rate-limited.
- One place for many tools. If you use Claude Code and Codex side by side, their providers, MCP servers, skills and prompts are managed together.
3. How to install CC Switch (and avoid fakes)
The official project is farion1231/cc-switch on GitHub, and
its README names ccswitch.io as the only official website. Search results
also show forks and mirrors — download only from the official repository's Releases
page or a package manager:
# macOS (signed and notarized build)
brew install --cask cc-switch
# Arch Linux
paru -S cc-switch-bin- Windows: the
.msiinstaller or portable.zipfrom Releases (Windows 10 and later). - Linux:
.deb,.rpmor.AppImagefrom Releases; needs glibc 2.35+ and WebKitGTK 4.1.
The app keeps its own data in ~/.cc-switch and backs up each tool's config
file before it rewrites it for the first time.
4. Using CC Switch with Claude Code
- Install Claude Code first, if you have not already.
- Add a provider in CC Switch: pick a preset for your provider (or add a custom one with its endpoint), and paste your API key from that provider.
- Choose the model the provider offers.
- Switch to the new provider in the main window or the tray.
- Start
claudein your project and check: ask it which model it is, or watch the provider's usage dashboard.
Under the hood, this is the same as setting Claude Code's environment yourself — a base URL pointing at a provider with an Anthropic-compatible API, a token and a model name. CC Switch just keeps those sets in one place, so you do not have to remember them.
If something does not switch, look for an override: a project's
.claude/settings.json or variables exported in your shell can win over what
CC Switch writes.
5. Using CC Switch with Codex
For Codex, the flow is the same: add the provider, switch, and restart Codex so it picks up the new configuration. One difference matters: providers speak different API formats. Where a provider's API does not match what the tool expects, CC Switch can run a local routing proxy that converts between formats — and the same proxy can fail over to a backup provider automatically.
6. MCP servers, skills and prompts in one place
Beyond providers, CC Switch manages the other things agents read:
- MCP servers — define a server once and enable it for the tools you want. If you are new to MCP, start with What is an MCP server.
- Skills — the same Agent Skills for several agents.
- Prompts — the instruction files each tool reads: CLAUDE.md for Claude, AGENTS.md for Codex.
It also shows usage and quota per provider and can sync its settings across machines through WebDAV or S3-compatible storage.
7. What to watch out for
- Where your code goes. A coding agent sends your source code to the model. Switching providers means a different company receives it — read their data policy. Be especially careful with third-party “relay” services that resell access to models; several are advertised around the project, and they see everything you send.
- Terms of service. The CC Switch README itself warns that using a subscription outside the vendor's official client may violate the vendor's terms. That risk is yours.
- Features tied to Anthropic. With Claude Code pointed at another provider, features that need Anthropic's API and a claude.ai account — such as Remote Control and cloud sessions — will not work. See Claude Code on mobile.
- Different models, different results. Claude Code's prompts and tools are tuned for Claude models. Other models can handle them well or poorly — test on a real task before you switch for good.
- Keys in one place. CC Switch stores all your provider keys locally. Protect the machine accordingly.
8. Alternatives to CC Switch
- Environment variables or shell aliases — a few lines in your shell profile per provider. No extra app, but easy to lose track of.
- OpenCode — an agent that is model-agnostic by design; switching models is built in. Compared in Claude Code vs Codex vs OpenCode.
- Local models — if privacy is the reason, run the model yourself; see Best local LLM for coding.
Do you see it differently? Write to support@coden-app.com.