Claude Max API Proxy: OpenAI-Compatible Subscription Endpoint

claude-max-api-proxy is an npm package maintained by the community (it is not an OpenClaw plugin). It presents a Claude Max/Pro subscription as an endpoint that speaks the OpenAI API, letting you aim any OpenAI-compatible tool at your subscription rather than at an Anthropic API key.

Warning

This is a technical compatibility shim, not a path Anthropic officially endorses. Anthropic has previously blocked certain subscription usage outside of Claude Code, so check Anthropic's present billing rules before you depend on this.

Anthropic's Claude Code documentation treats claude -p as Agent SDK/programmatic usage. Per Anthropic's June 15, 2026 support update, the Claude Agent SDK, claude -p, and third-party app usage all consume the signed-in subscription's usage limits (the separate Agent SDK credit plan announced earlier is paused). Consult Anthropic's Agent SDK plan article, the Pro/Max and Team/Enterprise plan articles, plus Anthropic provider for OpenClaw's own notes on Claude CLI billing.

Why use this

With this proxy, a Claude Max or Pro subscription can serve OpenAI-compatible tools. It is not an unlimited flat-rate arrangement: Claude Code's usage limits carry over to it. For production work, API keys stay the more predictable billing option.

How it works

Your App -> claude-max-api-proxy -> Claude Code CLI / claude -p -> Anthropic
     (OpenAI format)                (converts format)              (uses your login)

Each request causes the proxy to launch the Claude Code CLI as a subprocess. It translates OpenAI-format chat requests into CLI prompts, then streams back (or returns) the reply in OpenAI format.

Getting started

Install the proxy

Node.js 20+ is needed, along with a Claude Code CLI that has been authenticated.

npm install -g claude-max-api-proxy

# Verify Claude CLI is authenticated
claude --version
claude auth login   # if not already authenticated

Start the server

claude-max-api
# Server runs at http://localhost:3456

Test the proxy

curl http://localhost:3456/health
curl http://localhost:3456/v1/models

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

Configure OpenClaw

Configure OpenClaw to treat the proxy as a custom OpenAI-compatible endpoint:

{
  env: {
    vars: {
      OPENAI_API_KEY: "not-needed",
      OPENAI_BASE_URL: "http://localhost:3456/v1",
    },
  },
  agents: {
    defaults: {
      model: { primary: "openai/claude-opus-4" },
    },
  },
}

Note

The model ids listed below come from the proxy's own catalog and are not OpenClaw's Anthropic model refs. Every id corresponds to a Claude Code CLI model alias (opus, sonnet, haiku), which means the model behind it changes any time Anthropic updates that alias in the CLI. Look at the proxy's current README before depending on any particular mapping.

Advanced configuration

Proxy-style OpenAI-compatible notes

This relies on OpenClaw's generic custom /v1 OpenAI-compatible route, which is identical to the path used by any other self-hosted OpenAI-compatible backend:

• Request shaping that is native to OpenAI only does not apply.
• /fast and service_tier apply solely to direct api.anthropic.com traffic; proxy routes leave service_tier alone (refer to Anthropic provider fast mode).
• There is no Responses store, no prompt-cache hints, and no OpenAI reasoning-compat payload shaping.
• OpenClaw's OpenAI/Codex attribution headers (originator, version, User-Agent) go out only on native api.openai.com OAuth traffic. They are never sent to custom OPENAI_BASE_URL targets such as this proxy.

Auto-start on macOS with LaunchAgent

cat > ~/Library/LaunchAgents/com.claude-max-api.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-api</string>
  <key>RunAtLoad</key>
  <true/>
  <key>KeepAlive</key>
  <true/>
  <key>ProgramArguments</key>
  <array>
    <string>/usr/local/bin/node</string>
    <string>/usr/local/lib/node_modules/claude-max-api-proxy/dist/server/standalone.js</string>
  </array>
  <key>EnvironmentVariables</key>
  <dict>
    <key>PATH</key>
    <string>/usr/local/bin:/opt/homebrew/bin:~/.local/bin:/usr/bin:/bin</string>
  </dict>
</dict>
</plist>
EOF

launchctl bootstrap gui/$(id -u) ~/Library/LaunchAgents/com.claude-max-api.plist

Notes

• Claude Code's claude -p billing, usage-credit, and rate-limit behavior carries over.
• The listener binds to 127.0.0.1 only, and nothing is sent to a third-party server apart from the CLI's own call to Anthropic.
• Streaming responses are supported.
• Auth failures are not verified at startup. They appear only when a chat request actually runs, so an unauthenticated CLI means the first request fails instead of the server declining to start.

Note

If you want native Anthropic integration through Claude CLI or API keys, look at Anthropic provider. For OpenAI/Codex subscriptions, look at OpenAI provider.

Related

• Anthropic provider, Native OpenClaw integration with Claude CLI or API keys.

• OpenAI provider, For OpenAI/Codex subscriptions.

• Model selection, Overview of all providers, model refs, and failover behavior.

• Configuration, Full config reference.