OpenAI-compatible API proxy for Command Code. Use your Command Code subscription from any OpenAI-compatible client — OpenCode, Claude Code, or plain curl.

Command Code exposes two API surfaces:

This proxy talks /alpha/generate upstream and standard OpenAI downstream — so your existing plan works from any tool.

# Clone & run from source
git clone https://github.com/thaolaptrinh/commandcode-api-proxy.git
cd commandcode-api-proxy
npm install
npm run build && npm start

# Or run directly (no install)
npx commandcode-api-proxy

# Or install globally
npm install -g commandcode-api-proxy
commandcode-api-proxy

On first run, the proxy prompts for your Command Code API key (get it from https://commandcode.ai/settings). Other ways to provide it are in Authentication.

Provide your API key via --api-key, the CC_API_KEY env var, or save it with auth login (stored at ~/.config/commandcode-api-proxy/auth.json).

From source, replace commandcode-api-proxy with npm run auth -- (e.g. npm run auth -- login).

# Save a new API key
commandcode-api-proxy auth login

# Overwrite existing key
commandcode-api-proxy auth login --force

# Remove saved key
commandcode-api-proxy auth logout

Equivalent env vars (lower priority than CLI flags):

Security: the proxy forwards your paid Command Code key upstream and accepts any auth token from clients (proxy-managed), so it is designed for localhost use (HOST=127.0.0.1). Do not bind it to 0.0.0.0 on an untrusted network without restricting CORS_ORIGIN and putting your own auth in front.

All endpoints accept any auth token (use proxy-managed) — the proxy injects your real Command Code key upstream.

Run setup to auto-generate the provider config at ~/.config/opencode/opencode.json:

npx commandcode-api-proxy --setup-opencode

All CC models are listed directly — pick the one you want from the model selector.

Or add a commandcode provider manually — point baseURL at the proxy and use any model ID from the Model aliases table:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "commandcode": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Command Code",
      "options": {
        "baseURL": "http://127.0.0.1:8787/v1",
        "apiKey": "proxy-managed"
      },
      "models": {
        "deepseek-v4-pro": { "name": "DeepSeek V4 Pro" }
      }
    }
  }
}

Claude Code only offers three tiers — sonnet, opus, haiku. The setup maps each tier to a Command Code model via env vars in claude-settings.json:

npx commandcode-api-proxy --setup-claude-code

If a settings file already exists, re-run with --force to overwrite it.

Edit those env vars in the settings file to point at other CC models. To force every claude-* request to a single model regardless of tier, set ANTHROPIC_DEFAULT_MODEL. Then run:

claude --settings ~/.config/commandcode-api-proxy/claude-settings.json

Or set an alias:

alias claude-proxy="claude --settings ~/.config/commandcode-api-proxy/claude-settings.json"
claude-proxy

Short names work in addition to full model IDs:

Any model ID is passed through as-is — the proxy does not validate against a fixed list.

Some models support a reasoning_effort (low | medium | high | xhigh | max), and each accepts a different subset. The proxy sends the closest valid level for the model (e.g. deepseek-v4-pro supports only high/max, so low is clipped up to high and max is reachable). Models without a discrete effort set ignore the value. Pass the effort via OpenAI's reasoning_effort, or Anthropic's thinking.budget_tokens (larger budget → higher effort).

MIT