claude-code-proxy

Claude Code Proxy is a Python-based translation layer that intercepts requests from the Claude Code CLI and forwards them to any OpenAI-compatible backend, exposing a `/v1/messages` endpoint that mimics the Anthropic API. Developers point Claude Code at the local server by setting `ANTHROPIC_BASE_URL` to `http://localhost:8082`, and the proxy converts those calls to OpenAI-format requests before routing them to a configured provider such as standard OpenAI, Azure OpenAI, or a local Ollama instance. Model routing is handled through three environment variables (`BIG_MODEL`, `MIDDLE_MODEL`, `SMALL_MODEL`) that map Claude tier names like opus, sonnet, and haiku to whichever models the target provider offers. The proxy preserves tool use, SSE streaming, and base64 image inputs during translation. A notable detail is the `CUSTOM_HEADER_` prefix convention, which lets operators inject arbitrary HTTP headers into outbound requests without code changes. The project suits developers who want to use the Claude Code workflow against non-Anthropic models, whether for cost reasons, provider preference, or local inference.

What people ask about claude-code-proxy