API Reference Style Claude Code API relay English quick start

Claude Code API relay: a practical guide to endpoint setup, headers, and smoke tests

If you are evaluating a Claude Code API relay for tooling, CI checks, or a lightweight integration path, the main question is not marketing—it is compatibility. A good relay should behave like an OpenAI-compatible relay, preserve normal request structure, and make verification simple. This guide focuses on criteria you can actually test, including a basic configuration using OPENAI_BASE_URL=https://59api.com/v1.

Open relay console Review API-compatible endpoint

Endpoint

POST #/v1/chat/completions

For teams comparing 国内直连Claude, API中转站, or Claude 转发API options, start by checking whether the relay accepts standard chat-completions traffic, returns stable JSON, and handles timeouts cleanly. The best test is a small request, not a long prompt.

Criteria to check

OpenAI-style request and response structure.
Clear auth handling with a bearer token.
Predictable latency for short prompts.
Simple error messages when headers or model names are wrong.
Compatibility with local tools that read OPENAI_BASE_URL.

Headers

Use the same shape you would use for an OpenAI-compatible relay.

Authorization: Bearer YOUR_API_KEY
Content-Type: application/json
User-Agent: your-tool/1.0

Smoke-test checklist

Send one short prompt like “Reply with OK.”
Confirm HTTP 200 and valid JSON.
Verify the assistant content is returned in the expected field.
Repeat once after a restart or network change.

Example

Minimal request body for a first pass test.

{
  "model": "claude-3.5-sonnet",
  "messages": [
    { "role": "user", "content": "Reply with OK." }
  ],
  "temperature": 0.2
}

Config example

This is the key variable many tools respect without extra code.

export OPENAI_BASE_URL=#/v1
export OPENAI_API_KEY=YOUR_API_KEY

How to validate the setup

Set OPENAI_BASE_URL and the API key in your shell or app config.
Run a single chat request from your CLI, SDK, or internal tool.
Check that the relay returns a normal assistant message without schema edits.
Increase the prompt slightly and confirm behavior stays stable.
Log the status code, latency, and response size for future comparison.

Tip: If a request fails, first inspect the base URL, auth header, and model name. Those three checks solve most startup issues.

Short FAQ

Is this meant for direct application use?

Yes, if your app already speaks an OpenAI-style API, a relay can reduce integration work. Always verify the endpoint with a short smoke test first.

What should I watch for during a first test?

Look for a clean 200 response, stable JSON fields, and consistent behavior across two or three repeated requests. That tells you more than a long demo.

Where do I start if I use a CLI or SDK?

Set OPENAI_BASE_URL=#/v1, add your API key, and point the tool at a small prompt. If the tool supports OpenAI-compatible relay endpoints, it should connect with minimal changes.