Introduction

Claude Code is an incredible AI-powered coding assistant that runs in your terminal. While it typically connects to Anthropic’s API, did you know you can use it with GitHub Hosted Anthropic Models instead? This is particularly useful if you have access to GitHub’s AI model.

In this post, I’ll show you how to proxy Claude Code requests to GitHub-hosted Anthropic models using litellm, an open-source proxy server that translates between different LLM API formats.

Prerequisites

Before getting started, you’ll need:

  • Claude Code installed (see installation instructions)
  • Python 3.11 or higher
  • Access to GitHub Models (available through GitHub Copilot)

How It Works

The setup uses litellm-proxy as a translation layer:

Claude Code → litellm-proxy (localhost:4000) → GitHub Models API

The proxy intercepts Claude API calls and forwards them to GitHub’s infrastructure while presenting responses in the format Claude Code expects.

The Setup Script

I’ve created a bash script that automates the entire setup process. The script handles three main tasks:

  1. Virtual Environment Management: Creates a Python 3.11 venv and installs litellm
  2. Claude Configuration: Configures Claude Code to use the local proxy
  3. LiteLLM Configuration: Sets up the proxy to forward requests to GitHub Models

You can find the complete script here: litellm-proxy.sh

Key Configuration Details

The script modifies ~/.claude/settings.json to point Claude Code to the local proxy:

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "sk-",
    "ANTHROPIC_BASE_URL": "http://localhost:4000",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-41",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4.5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-sonnet-4.5",
    "ANTHROPIC_MODEL": "claude-sonnet-4.5",
    "ANTHROPIC_SMALL_FAST_MODEL": "claude-sonnet-4.5",
    "CLAUDE_CODE_SUBAGENT_MODEL": "claude-sonnet-4.5"
  }
}

It also creates a litellm configuration at ~/.config/litellm/config.yaml:

general_settings:
  master_key: sk-
litellm_settings:
  disable_copilot_system_to_assistant: true
  drop_params: true
model_list:
- model_name: '*'
  litellm_params:
    model: github_copilot/*
    extra_headers:
      Editor-Version: vscode/1.372.0

The wildcard model matching ('*') ensures all Claude model requests are caught and forwarded to GitHub’s backend. The Editor-Version header mimics VS Code to satisfy GitHub’s API requirements.

Usage

Starting the Proxy

First, start the litellm proxy server:

litellm-proxy.sh start

This will:

  • Create the Python virtual environment (on first run)
  • Install litellm and dependencies
  • Generate configuration files
  • Start the proxy server on http://localhost:4000

You should see output indicating the server is running:

➜  litellm-proxy.sh start
ℹ️ Activating virtualenv in /tmp/litellm
ℹ️ Creating claude settings file at /Users/suraj/.claude/settings.json
ℹ️ Creating litellm config file at /Users/suraj/.config/litellm/config.yaml
ℹ️ Starting litellm proxy server
INFO:     Started server process [94522]
INFO:     Waiting for application startup.

   ██╗     ██╗████████╗███████╗██╗     ██╗     ███╗   ███╗
   ██║     ██║╚══██╔══╝██╔════╝██║     ██║     ████╗ ████║
   ██║     ██║   ██║   █████╗  ██║     ██║     ██╔████╔██║
   ██║     ██║   ██║   ██╔══╝  ██║     ██║     ██║╚██╔╝██║
   ███████╗██║   ██║   ███████╗███████╗███████╗██║ ╚═╝ ██║
   ╚══════╝╚═╝   ╚═╝   ╚══════╝╚══════╝╚══════╝╚═╝     ╚═╝

...
INFO:     Uvicorn running on http://0.0.0.0:4000 (Press CTRL+C to quit)
...

Running Claude Code

In a separate terminal window, launch Claude Code as usual:

claude

Claude Code will now connect to the litellm proxy instead of Anthropic’s API, and your requests will be routed to GitHub Models!

Stopping and Cleanup

When you’re done, you can stop the proxy with Ctrl+C. The script automatically resets your Claude settings on exit.

If you want to manually reset Claude’s configuration:

litellm-proxy.sh reset-claude

To completely remove all configurations and the virtual environment:

litellm-proxy.sh cleanup

Why Use This Setup?

There are several compelling reasons to use Claude Code with GitHub Models:

  1. Cost Efficiency: GitHub Models offers a free tier for testing and development
  2. Existing Infrastructure: Leverage your existing GitHub subscriptions
  3. Rate Limits: Different rate limits compared to direct Anthropic API access
  4. Experimentation: Test Claude Code without committing to Anthropic API costs

Conclusion

Using Claude Code with GitHub-hosted Anthropic models is a great way to not pay for the $20/month subscription, especially if you already have access to the GitHub models. The litellm-proxy setup makes this seamless with just a simple script.

Give it a try and let me know how it works for you! You can find the complete script and other utilities in my dotfiles repository.

Resources