Command Code Proxy

中文文档

A reverse proxy that converts Command Code API to OpenAI / Anthropic compatible endpoints. Node.js ESM with zero external dependencies.

Built by analyzing the local command-code@1.15.1 CLI bundle and aligning the Command Code API request protocol.

Features: Native Command Code HTTP/WebSocket pass-through | OpenAI Chat Completions + Anthropic Messages API | Streaming & non-streaming | Tool calling (tool_use) | Multimodal image input | Reasoning effort | Dynamic model list | Cache hit metrics | Client disconnect detection with upstream abort | Zero-output → 429 auto-retry | Consecutive timeout → 429 auto-retry | Privacy-aware logging

Community: Linux.do — a friendly Chinese tech community.

npm start        # Start (default http://0.0.0.0:3050)
npm run dev      # Watch mode (auto-reload on file changes)

API Key is passed via the Authorization request header — no need to store it in config files. Key must start with user_ (automatically matched with any prefix, e.g. Bearer token_user_xxx):

curl http://127.0.0.1:3050/v1/chat/completions \
  -H "Authorization: Bearer user_xxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek/deepseek-v4-flash","messages":[{"role":"user","content":"hi"}]}'

commandcode/
├── config.json         # Port / log path etc.
├── LICENSE             # MIT License
├── package.json        # npm start / npm run dev
├── proxy.mjs           # HTTP entrypoint and request orchestration
├── src/                # Configuration, validation, state and CC client modules
├── test/               # Built-in node:test integration tests
├── Dockerfile          # Container build (node:22-alpine)
├── docker-compose.yml  # Container orchestration
├── .dockerignore       # Build context exclusions
├── README.md           # This document (English)
└── README_zh.md        # Chinese documentation

OpenAI Chat Completions compatible. Supports streaming, non-streaming, tool calling, multimodal image input, and reasoning effort.

Request parameters:

Simple request:

{
  "model": "deepseek/deepseek-v4-flash",
  "messages": [{ "role": "user", "content": "hello" }],
  "stream": true
}

Multimodal image input (vision model required):

{
  "model": "xiaomi/mimo-v2.5",
  "messages": [{
    "role": "user",
    "content": [
      { "type": "text", "text": "Describe this image" },
      { "type": "image_url", "image_url": { "url": "data:image/jpeg;base64,..." } }
    ]
  }]
}

Tool calling:

{
  "model": "deepseek/deepseek-v4-flash",
  "messages": [...],
  "tools": [{
    "type": "function",
    "function": { "name": "get_weather", "description": "...", "parameters": {...} }
  }],
  "tool_choice": "auto"
}

Streaming response (SSE):

data: {"id":"chatcmpl-xxx","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"thinking..."}}]}

data: {"id":"chatcmpl-xxx","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Hello"}}]}

data: {"id":"chatcmpl-xxx","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}],"usage":{"prompt_tokens":10,"completion_tokens":20,"total_tokens":30,"prompt_tokens_details":{"cached_tokens":8}}}

data: [DONE]

Non-streaming response (with cache hits):

{
  "id": "chatcmpl-xxx",
  "object": "chat.completion",
  "created": 1234567890,
  "model": "deepseek/deepseek-v4-flash",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "Hello!",
      "reasoning_content": "The user said hello, I should respond."
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 7558,
    "completion_tokens": 42,
    "total_tokens": 7600,
    "prompt_tokens_details": { "cached_tokens": 7552 }
  }
}

Anthropic Messages API compatible endpoint. Supports streaming, non-streaming, and tool calling.

Request body:

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1000,
  "system": "You are a helpful assistant.",
  "messages": [
    { "role": "user", "content": "hello" }
  ],
  "stream": true
}

Anthropic protocol conversion (automatic):

Streaming response (SSE, Anthropic format):

event: message_start
data: {"type":"message_start","message":{"id":"msg_xxx","type":"message","role":"assistant","content":[],"model":"...","usage":{"input_tokens":0,"output_tokens":0}}}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Hello"}}

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":10,"cache_read_input_tokens":0,"input_tokens":100}}

event: message_stop
data: {"type":"message_stop"}

Non-streaming response:

{
  "id": "msg_xxx",
  "type": "message",
  "role": "assistant",
  "model": "deepseek/deepseek-v4-flash",
  "content": [{ "type": "text", "text": "Hello!" }],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 7558,
    "output_tokens": 42,
    "cache_read_input_tokens": 7552,
    "cache_creation_input_tokens": null
  }
}

Returns the official Provider API model list (5 min cache); the upstream endpoint supports anonymous access, and upstream failures return an error instead of a local fallback list.

Health check. Returns OK.

The proxy forwards native Command Code API requests on the same path. /alpha/*, /provider/*, and the /beta/* and /internal/* namespaces declared by the 1.15.1 bundle are sent only to the configured apiBase. Other paths are not forwarded, so this is not an arbitrary URL proxy.

Native requests preserve the HTTP method, query, raw body bytes, authentication/OAuth/cookie headers, upstream status, response headers, and response stream. Only hop-by-hop headers such as Host, Connection, and Transfer-Encoding are removed; 3xx responses are not followed automatically. The 10MB request limit still applies. Use a dedicated hostname for the native endpoint in production so it does not share cookies with unrelated web applications.

curl http://127.0.0.1:3050/alpha/whoami \
  -H "Authorization: Bearer user_xxxxxxxxx" \
  -H "x-command-code-version: 1.15.1"

POST /alpha/generate returns the native newline-delimited JSON stream without converting it to OpenAI SSE. Sandbox real-time connections use a WebSocket tunnel on the same path, such as ws://127.0.0.1:3050/alpha/sandbox/stream/.... External OAuth, npm updates, telemetry, and user-configured MCP origins are outside the Command API origin and are not proxied by this endpoint.

The proxy returns a live model list via GET /v1/models. Below are common models for reference; the actual list depends on the live API response — see Command Code Pricing for plan details.

The proxy returns the official Provider API JSON without constructing a local model list. The upstream endpoint supports anonymous access; upstream failures or disabled Provider API return an error instead of fabricated model data. The response is cached per API key or publicly for 5 minutes.

⚠️ Some models (e.g. deepseek-v4-flash, claude-sonnet-4-6) do not support image input. Use xiaomi/mimo-v2.5, Kimi-K2.5, or other vision models for multimodal.

from openai import OpenAI

client = OpenAI(
    api_key="user_xxxxxxxxx",
    base_url="http://127.0.0.1:3050/v1",
)

response = client.chat.completions.create(
    model="deepseek/deepseek-v4-flash",
    messages=[{"role": "user", "content": "hello"}],
    stream=True,
)
for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")

curl http://127.0.0.1:3050/v1/chat/completions \
  -H "Authorization: Bearer user_xxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash",
    "messages": [{"role": "user", "content": "hello"}],
    "stream": true
  }'

Add a Custom Provider in Cursor settings:

• API Base URL: http://127.0.0.1:3050/v1
• API Key: user_xxxxxxxxx
• Model: Choose from the model list

import anthropic

client = anthropic.Anthropic(
    api_key="user_xxxxxxxxx",
    base_url="http://127.0.0.1:3050",
)
message = client.messages.create(
    model="deepseek/deepseek-v4-flash",
    max_tokens=1000,
    system="You are helpful.",
    messages=[{"role": "user", "content": "hello"}],
)
print(message.content[0].text)

{
  "provider": "openai-compatible",
  "baseUrl": "http://127.0.0.1:3050/v1",
  "apiKey": "user_xxxxxxxxx"
}

Based on analysis of the local command-code@1.15.1 bundle:

{
  "config": {
    "workingDir": "C:\\project",
    "date": "2026-06-07",
    "environment": "linux",
    "structure": [],
    "isGitRepo": false,
    "currentBranch": "",
    "mainBranch": "",
    "gitStatus": "",
    "recentCommits": []
  },
  "memory": null,
  "taste": null,
  "skills": null,
  "permissionMode": "standard",
  "mode": "agent",
  "params": {
    "model": "deepseek/deepseek-v4-flash",
    "messages": [...],
    "tools": [],
    "max_tokens": 64000,
    "stream": true,
    "reasoning_effort": "max"
  },
  "threadId": "<uuid>"
}

threadId is sent only when the client provides a valid UUID through x-thread-id or x-command-code-thread-id; otherwise it is omitted as in the latest CLI.

The CLI sends images in this format:

{
  "role": "user",
  "content": [
    { "type": "image", "image": "data:image/jpeg;base64,..." },
    { "type": "text", "text": "What does this image say?" }
  ]
}

The proxy receives OpenAI image_url format and converts it to the above CC format transparently.

docker compose up -d

The proxy will listen on http://0.0.0.0:3050. Set PROXY_PORT to customize the host port:

PROXY_PORT=13050 docker compose up -d

docker build -t commandcode-proxy:latest .
docker run -d -p 3050:3050 -e PORT=3050 commandcode-proxy:latest

npm run docker:build:multi

This project is for educational and research purposes only.

• Unofficial: This project is not affiliated with Command Code in any way.
• Personal Use: Users assume all responsibility. Please comply with the Command Code Terms of Service.
• API Key: This project does not collect, upload, or leak your API Key. The key must be sent in every request via the Authorization: Bearer <key> header and is never stored in configuration.
• Compliance: The protocol is based on passive observation of local CLI network traffic. No unauthorized access, cracking, or tampering of the server has been performed.
• Account Risk: Keep usage frequency consistent with normal CLI usage. Extremely high concurrent calls may trigger risk controls.

# Start with watch mode (auto-reload on file changes)
npm run dev

# Run syntax checks and built-in integration tests
npm run check
npm test