free-claude-code: a local proxy that lets Claude Code use many model providers

Claude Code is built around Anthropic’s models. If you like its interface and workflow but do not want a second bill from Anthropic, you need a layer in between. free-claude-code is an AGPL-3.0 open-source project that runs a proxy on your machine and forwards requests from Claude Code, Codex, VS Code and other clients to the model providers you configure. The README says it works with 61 providers, covering free tiers, paid APIs, subscription accounts and local models. It also states that the project is independent and is not affiliated with or endorsed by Anthropic.

The value is in switching. The same client can point at another model, or fall back to a backup model when a provider throttles or goes down, without restarting the turn. That also means quality, tool-use reliability and cost depend on the provider you choose. The README does not compare coding quality across models, and we did not run such a comparison.

Core features

• One interface for many providers: Free, paid, subscription and local models are configured in one searchable admin UI. The provider catalog includes Anthropic, which uses a separately billed API key, along with OpenAI API, GitHub Copilot, Vercel AI Gateway and Hugging Face Inference Providers.
• Automatic fallback: When the retries for one turn run out, FCC moves to the next model in your list, and the turn continues without a restart. The README says this works with every client.
• Several clients, one proxy: Claude Code in VS Code, VS Code Chat, Claude Desktop, Codex in VS Code and its app, and JetBrains through ACP can all connect. Fully quit Claude Desktop before you connect it, then reopen it.
• Subscriptions through OAuth: A ChatGPT subscription connects through the OAuth providers page, and GitHub Copilot uses the GitHub account you sign in with. Anthropic uses a separately billed API key.
• Voice and chat entry points: Voice notes can be transcribed with local Whisper or NVIDIA NIM, and messages can come in through Discord.
• Codex in the browser: Pick a folder, run Codex in the browser, and switch providers and models inside the same session.
• Proxy authentication: Turn on Proxy Authentication to require a bearer token for the local proxy.

Typical use cases

• A developer who likes Claude Code’s interface but does not want Anthropic billing: Try small tasks on a provider’s free tier, and pay only once the workflow proves itself. Free tiers belong to each provider, and the README says they may change.
• Someone who wants a second model to review code: Run the same task in the same repository with another model, then compare the two outputs.
• A team that does not trust one provider’s uptime: Configure two or three backup models for the same client, so a throttled or failed provider does not stop the work.
• Developers who work from an IDE or a phone: VS Code, JetBrains and the voice entry points share one configuration.

If your main concern is keeping long-running agent work going across sessions rather than switching models, compare it with LoopX.

Quick start

On macOS and Linux, use the one-line installer from the README:

curl -fsSL "https://raw.githubusercontent.com/Alishahryar1/free-claude-code/main/scripts/install.sh" | sh

On Windows, run the PowerShell command in the README, which fetches install.ps1 from the same repository. Piping a remote script straight into sh runs whatever the server returns, so download it and read it before you run it.

After the install:

1. Open the admin UI, go to Providers, choose a provider, paste its API key and save. For NVIDIA NIM, the README’s steps are Add key, paste the key, then Save.
2. Pick a model from the MODEL dropdown. If a provider cannot list its models, type <provider-id>/<exact-provider-model-id> by hand.
3. Connect your client to FCC, then start a new session.

Summary

free-claude-code suits people who already use Claude Code, Codex or VS Code’s AI features and want to switch models, keep a backup, or manage cost. It does not suit anyone who expects Anthropic support or a service guarantee.

A few things to know first. Free-tier availability and limits belong to each provider, and the README says they may change. Subscription connections, such as ChatGPT and GitHub Copilot, fall under each provider’s own terms. The README says FCC follows provider terms and removes an integration when a provider stops allowing it, but we did not check each provider’s terms one by one. The license is AGPL-3.0-only. If you modify the software and offer the modified version to people over a network, AGPL section 13 requires you to offer those users the source of your modified version.

We checked the README, the LICENSE file and the repository metadata only. We did not install or run the tool, and we did not compare coding quality across models. GitHub’s license detector returns NOASSERTION for this repository, so we took the license from the LICENSE file. As of October 21, 2026, the repository had 62,273 stars, its last push was on October 18, 2026, and it is not archived.