claude-max-openai-proxy

Use your Claude Max/Pro subscription as an OpenAI-compatible API — with full tool support.

Unlike other proxies that wrap the Claude Code CLI in --print mode (text-only), this proxy uses the official Claude Agent SDK which gives Claude access to real tools — file I/O, shell commands, web search, and more. Claude runs your tasks, not just answers your questions.

Why?

Feature	claude-max-api-proxy	This proxy
OpenAI-compatible API	✅	✅
Uses Claude Max subscription	✅	✅
Tool calling (file read/write)	❌	✅
Shell command execution	❌	✅
Web search & fetch	❌	✅
Official SDK	❌ (CLI hack)	✅ (claude-agent-sdk)
Streaming	✅	✅

Quick Start

Prerequisites

1. Node.js 18+
2. Claude Code CLI — authenticated with your Claude Max or Pro subscription:

   npm install -g @anthropic-ai/claude-code
   claude auth login

3. Claude Agent SDK:

   npm install -g @anthropic-ai/claude-agent-sdk

Install & Run

# Clone the repo
git clone https://github.com/0x0ndra/claude-max-openai-proxy.git
cd claude-max-openai-proxy

# Start the proxy
node server.mjs

The proxy starts on http://127.0.0.1:3456 by default.

Test It

curl -X POST http://localhost:3456/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4",
    "messages": [{"role": "user", "content": "What files are in the current directory?"}]
  }'

Claude will actually run ls and tell you what's there. That's the difference.

Configuration

All configuration is via environment variables:

Variable	Default	Description
PORT	3456	Port to listen on
HOST	127.0.0.1	Host to bind to (use 0.0.0.0 for all interfaces)
PROXY_CWD	Current directory	Working directory for Claude's tools
MAX_TURNS	15	Max tool-calling rounds per request
ALLOWED_TOOLS	Bash,Read,Write,Edit,Glob,Grep,WebFetch,WebSearch	Comma-separated list of allowed tools

Example with custom config:

PORT=8080 MAX_TURNS=5 ALLOWED_TOOLS="Read,Glob,Grep,WebSearch" node server.mjs

API Endpoints

POST /v1/chat/completions

OpenAI-compatible chat completions endpoint. Supports:

• messages — array of {role, content} messages
• model — model name (see below)
• stream — true for SSE streaming
• max_turns — override max tool-calling rounds

GET /v1/models

Returns available models.

GET /health

Health check with status, tools, and model info.

Available Models

Model name	Maps to
claude-opus-4	claude-opus-4-20250514
claude-sonnet-4	claude-sonnet-4-5-20250929
claude-sonnet-4-5	claude-sonnet-4-5-20250929
claude-haiku-4	claude-haiku-4-5-20251001
claude-haiku-4-5	claude-haiku-4-5-20251001

You can also use the full model ID directly.

Available Tools

Claude can use these tools internally while processing your request:

Tool	Description
Bash	Execute shell commands
Read	Read file contents
Write	Create/overwrite files
Edit	Edit existing files (search & replace)
Glob	Find files by pattern
Grep	Search file contents
WebFetch	Fetch and process web pages
WebSearch	Search the web

Tools run inside the proxy — your client just gets the final text response. No need to implement the OpenAI tool-calling protocol on your side.

Use with OpenClaw

This proxy was built to power OpenClaw bots using a Claude Max subscription instead of API credits. Config example:

{
  "models": {
    "providers": {
      "claudemax": {
        "baseUrl": "http://localhost:3456/v1",
        "apiKey": "not-needed",
        "api": "openai-completions",
        "models": [
          { "id": "claude-opus-4", "name": "Claude Opus 4", "contextWindow": 200000, "maxTokens": 8192 },
          { "id": "claude-sonnet-4", "name": "Claude Sonnet 4", "contextWindow": 200000, "maxTokens": 8192 }
        ]
      }
    }
  },
  "agents": {
    "defaults": { "model": { "primary": "claudemax/claude-opus-4" } },
    "list": [{
      "id": "personal",
      "tools": { "byProvider": { "claudemax": { "deny": ["*"] } } }
    }]
  }
}

Note: Deny OpenClaw's built-in tools for the claudemax provider — the proxy handles tools internally via Claude Code's own toolset.

Run as macOS Service

Create a LaunchDaemon for auto-start:

sudo tee /Library/LaunchDaemons/com.claude-max-openai-proxy.plist << EOF
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>Label</key>
    <string>com.claude-max-openai-proxy</string>
    <key>UserName</key>
    <string>$(whoami)</string>
    <key>ProgramArguments</key>
    <array>
        <string>$(which node)</string>
        <string>$(pwd)/server.mjs</string>
    </array>
    <key>EnvironmentVariables</key>
    <dict>
        <key>PATH</key>
        <string>$(dirname $(which claude)):$(dirname $(which node)):/usr/bin:/bin</string>
        <key>HOME</key>
        <string>$HOME</string>
    </dict>
    <key>RunAtLoad</key>
    <true/>
    <key>KeepAlive</key>
    <true/>
</dict>
</plist>
EOF

sudo launchctl load /Library/LaunchDaemons/com.claude-max-openai-proxy.plist

How It Works

Your App  ──POST /v1/chat/completions──▶  Proxy  ──query()──▶  Claude Agent SDK
                                                                      │
                                                              ┌───────┴───────┐
                                                              │  Claude Code  │
                                                              │  subprocess   │
                                                              └───────┬───────┘
                                                                      │
                                                           Uses Claude Max/Pro
                                                           subscription (OAuth)
                                                                      │
                                                              ┌───────┴───────┐
                                                              │   Claude AI   │
                                                              │  (Anthropic)  │
                                                              └───────┬───────┘
                                                                      │
                                                            Tool calls handled
                                                            internally (Bash,
                                                            Read, Write, etc.)
                                                                      │
Your App  ◀──── Final text response ────  Proxy  ◀── result ─────────┘

1. Your app sends an OpenAI-format request
2. The proxy converts messages to a prompt and calls query() from the Claude Agent SDK
3. The SDK spawns Claude Code as a subprocess (authenticated with your subscription)
4. Claude processes the request, using tools as needed (reading files, running commands, etc.)
5. The proxy collects Claude's final text response and returns it in OpenAI format

Security

• Bind to localhost by default — the proxy is not exposed to the network
• Claude runs tools with your user's permissions — be careful with Bash and Write tools
• No API keys needed — authentication is handled by Claude Code's OAuth
• To restrict tools, set ALLOWED_TOOLS=Read,Grep,WebSearch (no write/exec access)

Limitations

• No OpenAI-style tool calling protocol — tools run inside the proxy, not via tool_calls responses. Your client gets the final text only.
• Token usage not tracked — the usage field in responses shows zeros (Claude Code doesn't expose token counts via the SDK in a usable format for the OpenAI schema).
• One request at a time — each request spawns a Claude Code subprocess. Concurrent requests work but may hit subscription rate limits.
• Claude Max/Pro subscription required — this is not a free API. You're using your own subscription.

License

MIT — see LICENSE.

Author

0x0ndra — github.com/0x0ndra