Is the Go proxy implementation persistent?

Yes, it includes patterns for macOS launchd configuration, ensuring the proxy runs as a daemon that persists across reboots and automatically restarts if it crashes.

Why would I use a proxy with Claude Code?

Using a proxy allows you to route different model tiers to different providers, such as using MiniMax for Haiku to save costs while keeping Sonnet on Anthropic's native infrastructure.

What is the 'proxy-managed' API key used for?

Setting ANTHROPIC_API_KEY to 'proxy-managed' acts as a sentinel that tells Claude Code to switch from OAuth-only mode to honoring the ANTHROPIC_BASE_URL setting.

How does this skill handle Claude's OAuth authentication?

It implements patterns to read OAuth tokens directly from the macOS Keychain ('Claude Code-credentials') and injects the required 'oauth-2025-04-20' beta headers into proxied requests.

Can I use this for non-Anthropic providers?

Yes, it is designed to work with any provider that supports the Anthropic /v1/messages API format, such as MiniMax or other compatible LLM gateways.

Claude Code Multi-Provider Proxy

Name: Claude Code Multi-Provider Proxy
Author: terrylica

byterrylica

•

部署与 DevOps

Optimizes Claude Code performance and cost by routing specific model tiers to different AI providers via a local reverse proxy.

This skill provides expert patterns for implementing a local Go-based reverse proxy that intelligently routes Claude Code requests based on model tier. It allows users to send lightweight Haiku requests to cost-effective providers like MiniMax while maintaining native Anthropic access for Sonnet and Opus models. The skill addresses complex authentication challenges by automating macOS Keychain OAuth token retrieval, managing required beta headers, and implementing robust launchd service configurations for background persistence and automated failover.

主要功能

01Advanced header management for mandatory 'oauth-2025-04-20' compatibility

02Built-in concurrency control with semaphores and exponential backoff retries

03Launchd service integration for persistent background operation and auto-restart

0424 GitHub stars

05Intelligent model routing between Anthropic and compatible providers like MiniMax

06Automated macOS Keychain OAuth token retrieval and secure caching

使用场景

01Reducing API costs by routing low-tier model requests to cheaper third-party providers

02Setting up a resilient multi-provider LLM environment with automated failover logic

03Resolving 'OAuth not supported' errors when using Claude Code behind a custom proxy

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add terrylica/cc-skills claude-code-proxy-patterns

For use in Claude.ai and ChatGPT

主要功能

01Advanced header management for mandatory 'oauth-2025-04-20' compatibility

02Built-in concurrency control with semaphores and exponential backoff retries

03Launchd service integration for persistent background operation and auto-restart

0424 GitHub stars

05Intelligent model routing between Anthropic and compatible providers like MiniMax

06Automated macOS Keychain OAuth token retrieval and secure caching

使用场景

01Reducing API costs by routing low-tier model requests to cheaper third-party providers

02Setting up a resilient multi-provider LLM environment with automated failover logic

03Resolving 'OAuth not supported' errors when using Claude Code behind a custom proxy

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add terrylica/cc-skills claude-code-proxy-patterns

For use in Claude.ai and ChatGPT