Open source · MIT licensed · Runs locally

Run every model you already pay for, right inside Claude Code

ccmux is a tiny local proxy that turns your GitHub Copilot subscription and your Ollama Cloud key into backends for Claude Code, so the model picker finally has everything you already own.

npm i -g @dondetir/ccmux && ccmux claude
No relay servers. Binds to 127.0.0.1 only. Built with a few hundred lines of TypeScript.

About ccmux

A small proxy with one job: give Claude Code access to every model you are already paying for.

Claude Code only ever spoke to one backend, even though most developers already have a GitHub Copilot subscription and maybe an Ollama Cloud key sitting unused. ccmux runs quietly on your own machine and translates between the Anthropic API that Claude Code expects and whatever backend you point it at.

It is an independent, community project, not an official Anthropic or GitHub product. It rides on Copilot’s editor-facing API rather than a public one, so it can break if that surface changes, and free-tier premium requests are still limited by Copilot itself. ccmux just makes sure those limits are handled gracefully.

Status

Actively maintained, unofficial, MIT licensed.

Footprint

A few hundred lines of TypeScript, no external services.

Requirements

Node 20+, Claude Code, and a GitHub account with Copilot access.

Don’t switch tools. Switch backends.

Your subscriptions already cover more models than Claude Code can normally see. ccmux just wires them up.

🧩

One picker, every backend

Copilot already ships Claude, GPT and Gemini. Ollama Cloud adds GLM, Qwen, DeepSeek and Kimi. ccmux lines every one of them up behind Claude Code’s /model command.

🔒

Nothing leaves your machine

The proxy binds to 127.0.0.1 only, so your tokens and traffic stay local with no third-party relay in the middle.

🆓

Plays nicely with free tiers

Built to work even on Copilot’s free plan. Rate limits are passed straight through so Claude Code backs off cleanly instead of erroring out.

How it works

ccmux speaks the Anthropic API on one side and the right backend API on the other, translating as needed.

Claude Code

your terminal
Anthropic API

ccmux

local proxy · :4141
Copilot API

GitHub Copilot

your subscription

Copilot’s own Claude models pass through natively. GPT, Gemini and every Ollama model get translated on the fly, and reasoning from thinking-capable Ollama models streams back as Claude-style thinking blocks.

Every model, one /model picker

No settings.json edits needed. They just show up.

claude-haiku-4.5
claude-sonnet-4.6
claude-gpt-4.1
claude-gemini-2.5-pro
ollama/glm-5.2
ollama/qwen3-coder:480b
ollama/deepseek
ollama/kimi

Free Copilot tier unlocks Haiku. A paid plan adds Sonnet, GPT and Gemini. Drop in an Ollama Cloud key to bring the rest along, each with its real context window.

Up and running in under a minute

Needs Node 20+, Claude Code, and a GitHub account with Copilot access.

1

Install

Puts the ccmux CLI on your PATH.

npm i -g @dondetir/ccmux
2

Launch

Starts the proxy if it isn’t running, then opens Claude Code.

ccmux claude
3

Sign in once

First run triggers a one-time device-flow login, saved locally for next time.

~/.config/ccmux/env