claude-cli-proxy

Lightweight OpenAI-compatible proxy for Claude Code CLI. Zero npm dependencies.

Claude Max/Pro subscribers get unlimited Claude access via the CLI but no API key. This proxy wraps claude -p as an HTTP server that speaks the OpenAI API format, so you can use any OpenAI-compatible client with your Claude subscription.

Why not claude-max-api-proxy?

The existing claude-max-api-proxy spawns claude -p without isolating the subprocess from your user config. Each request loads all your hooks, plugins, and MCP servers — causing infinite loops and massive token waste.

claude-cli-proxy fixes this with clean subprocess flags:

claude -p \
  --setting-sources ""      # skip ALL settings (hooks, plugins)
  --strict-mcp-config       # skip ALL MCP servers
  --tools ""                # no tool use
  --max-budget-usd 0.5      # safety cap per request

Quick Start

git clone https://github.com/SenZhangAI/claude-cli-proxy.git
cd claude-cli-proxy
node proxy.mjs

No npm install needed — zero dependencies.

Usage

Test with curl

# Non-streaming
curl http://localhost:3456/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-haiku-4","messages":[{"role":"user","content":"Hello"}]}'

# Streaming
curl -N http://localhost:3456/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4","messages":[{"role":"user","content":"Hello"}],"stream":true}'

Use with any OpenAI client

export OPENAI_BASE_URL=http://localhost:3456/v1
export OPENAI_API_KEY=not-needed

# Python
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model="claude-sonnet-4",
    messages=[{"role": "user", "content": "Hello"}]
)

# Node.js
import OpenAI from 'openai';
const client = new OpenAI();
const response = await client.chat.completions.create({
    model: 'claude-sonnet-4',
    messages: [{ role: 'user', content: 'Hello' }],
});

Configuration

Environment variables:

Variable	Default	Description
PORT	3456	Server port
DEFAULT_MODEL	sonnet	Model when none specified
TIMEOUT	120	Request timeout in seconds
MAX_CONCURRENT	3	Max simultaneous requests
MAX_BUDGET	0.5	Dollar cap per request

PORT=8080 DEFAULT_MODEL=haiku MAX_CONCURRENT=5 node proxy.mjs

Endpoints

Method	Path	Description
POST	/v1/chat/completions	OpenAI-compatible chat completions
GET	/v1/models	List available models
GET	/health	Health check with active request count

Available Models

OpenAI Model ID	Claude Model
claude-opus-4	Opus
claude-sonnet-4	Sonnet
claude-haiku-4	Haiku

Short aliases (opus, sonnet, haiku) also work.

Prerequisites

• Node.js 18+
• Claude Code CLI installed and authenticated (claude auth)

How It Works

1. Receives OpenAI-format request
2. Converts messages to a prompt string
3. Spawns claude -p with clean isolation flags
4. Pipes prompt via stdin, reads response from stdout
5. Returns OpenAI-format JSON (or SSE for streaming)

Subprocess cleanup is handled automatically — client disconnects kill the subprocess, timeouts are enforced, and all processes are cleaned up on server exit.

License

MIT