Claude OAuth Proxy + Anti-Ban Engine

A local proxy that uses your Claude Pro/Max subscription as a standard Anthropic API endpoint, with built-in anti-detection stealth layer.

Features

Core Proxy

• OAuth Authentication — Uses Claude setup-token for subscription access
• Full API Compatibility — Drop-in replacement for Anthropic API
• Streaming Support — Full SSE streaming passthrough
• PM2 Ready — Production-grade process management

🛡️ Anti-Ban Engine

Makes proxy traffic indistinguishable from real Claude Code CLI usage:

Layer	What it does
Human Timing	Random delays (0.6-3.2s), thinking pauses, coffee breaks, session patterns
Tool Injection	Injects real Claude Code tool definitions (Bash, Read, Write, Edit, Glob, Grep, etc.)
System Prompt	Rewrites system prompt to match Claude Code's exact format and structure
Header Fingerprint	Rotates CLI version, platform strings, includes all telemetry headers
Session Simulation	Coding bursts (3-10 req) → pause → burst. Sessions with natural breaks
Quiet Hours	Reduced activity during configurable sleep hours (default: 21:00-04:00 UTC)
Rate Governance	Per-minute (8), per-hour (100), per-day (600) limits matching human capacity

Quick Start

# 1. Generate a setup token (requires Claude Code CLI)
claude setup-token

# 2. Start the proxy
node proxy.mjs --token sk-ant-oat01-your-token-here

# 3. Use it
curl http://localhost:4321/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: anything" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Production Setup (PM2)

pm2 start ecosystem.config.js
pm2 save

Endpoints

Path	Description
GET /health	Proxy status + anti-ban stats
POST /v1/messages	Chat completions (streaming supported)
POST /v1/messages/count_tokens	Token counting

Configuration

Token Sources (checked in order)

1. --token CLI flag
2. CLAUDE_CODE_OAUTH_TOKEN environment variable
3. Saved token in ~/.config/claude-oauth-proxy/token
4. Interactive prompt

Environment Variables

Variable	Default	Description
PORT	4321	Proxy listen port
HOST	127.0.0.1	Bind address
CLAUDE_CODE_OAUTH_TOKEN	—	OAuth token

Anti-Ban Tuning

Edit the CONFIG object in anti-ban.mjs to adjust:

• Delay ranges and probabilities
• Rate limits per minute/hour/day
• Session burst size and break duration
• Quiet hours (timezone)
• Claude Code version pool

How It Works

Your App → Proxy (anti-ban gate) → Anthropic API
              ↓
    1. Rate limit check
    2. Session simulation delay
    3. Human thinking delay
    4. Tool definition injection
    5. System prompt rewrite
    6. Header fingerprint rotation
    7. Forward with OAuth Bearer auth

Every request that reaches Anthropic looks exactly like a real Claude Code CLI session:

• Correct user-agent and telemetry headers
• Full Claude Code tool definitions attached
• System prompt matches Claude Code's format
• Request timing follows human patterns

Security

• Binds to 127.0.0.1 only — not accessible from other machines
• Token stored with 0600 permissions
• No data logging — only method, path, status, and model name
• Client API keys are ignored (any value works, it's localhost)

Disclaimer

This uses the same OAuth client ID and headers as Claude Code. It is not officially sanctioned by Anthropic. They could change required headers, rotate client identity, or block this approach at any time.

The setup-token is valid for ~1 year but can be revoked. Always have a regular API key as a fallback.

License

MIT