ccr-codex-copilot-lmstudio

Curated setup for Claude Code Router with example configurations for:

• GitHub Copilot
• OpenAI Codex (Responses API)
• LM Studio (local OpenAI-compatible server)
• Kilo Code (kilo.ai)

What this is based on

This setup follows the same routing + transformer model as the upstream claude-code-router project, where:

• Providers define upstreams (Copilot, Codex, LM Studio).
• Transformers adapt request/response formats or inject headers.
• Router picks default provider/model pairs.
  See the upstream project for full context and updates: https://github.com/musistudio/claude-code-router

Contents

• config.example.json — example router config with Copilot/Codex/LM Studio.
• presets/example/manifest.json — preset template for quick switching.
• plugins/copilot-transformer.js — Copilot header injector + token refresh.
• plugins/codex-transformer.js — Chat Completions → Responses API conversion.
• plugins/kilo-transformer.js — Kilo Code header injector and request transformer.
• codex-headers.example.json — Codex auth header
• scripts/copilot-auth-simple.sh — device-flow login + token cache helper.

Quick start

1. Install or clone Claude Code Router (CCR).
2. Copy example files into your CCR directory:
   • config.example.json → config.json
   • codex-headers.example.json → codex-headers.json
3. Set CCR_HOME to your CCR install directory:
   • Example: export CCR_HOME="$HOME/.claude-code-router"
4. Edit config.json (see below) and start CCR.

Configuration notes

1) Providers

config.example.json includes three providers:

• LM Studio: http://localhost:1234/v1/chat/completions
• Copilot: https://api.individual.githubcopilot.com/chat/completions
• Codex: https://chatgpt.com/backend-api/codex/responses

2) Transformers

• copilot-transformer:
  • Adds required Copilot headers.
  • Uses a local token cache (~/.copilot-tokens.json by default).
  • Auto-refreshes token when close to expiry.
• codex-transformer:
  • Converts Chat Completions → Responses API format.
  • Merges headers from codex-headers.json.
  • Can enable reasoning via options.reasoning.
• kilo-transformer:
  • Adds Kilo Code-specific headers (e.g., X-KiloCode-Version, X-KiloCode-TaskId).
  • Loads custom headers from an optional headers file.
  • Strips Authorization header unless overridden (configurable via strip_authorization).
  • Generates random task IDs for each request.

3) Codex headers

codex-headers.example.json is a template.
Create codex-headers.json and paste real values there.

Capturing Codex headers (VS Code)

Recommended approaches:

• Proxy the Extension Host: launch VS Code with HTTPS_PROXY/HTTP_PROXY set to a local proxy (Proxyman, Charles, mitmproxy) and trust the proxy cert. This shows the actual /backend-api/codex/responses request headers you need.
• OS-level capture: use Wireshark/tcpdump to inspect traffic from the Extension Host process.

Once you see the request, copy the relevant request headers into codex-headers.json.

4) Copilot auth

Run:

bash scripts/copilot-auth-simple.sh

This script will:

• authenticate via GitHub device flow
• fetch a Copilot token
• save it to ~/.copilot-tokens.json
• update your CCR config automatically (if jq is installed)

5) Presets

If you use CCR presets, copy presets/example/manifest.json and customize it.

Not affiliated

This repository is an independent setup helper. It is not affiliated with GitHub, OpenAI, or Anthropic.