claude-code-relay 0.0.7

Local proxy that exposes Claude CLI as an OpenAI-compatible API server

claude-code-relay

Local proxy that exposes Claude CLI as an OpenAI-compatible API server.

Use your existing Claude CLI installation with any OpenAI-compatible client.

Why?

• You already have Claude CLI working with your subscription
• You want to use tools that expect OpenAI API format
• No separate API key needed - uses your local Claude CLI

Installation

Node.js / Bun

npx claude-code-relay serve
# or
bunx claude-code-relay serve
# or install globally
npm install -g claude-code-relay

Python

pip install claude-code-relay

Usage

Start the server

# Node
npx claude-code-relay serve --port 52014

# Python
claude-code-relay serve --port 52014

Use with OpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:52014/v1",
    api_key="not-needed"
)

response = client.chat.completions.create(
    model="sonnet",  # or "opus", "haiku"
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "http://localhost:52014/v1",
  apiKey: "not-needed",
});

const response = await client.chat.completions.create({
  model: "sonnet",
  messages: [{ role: "user", content: "Hello!" }],
});

Use with LiteLLM

from litellm import completion

response = completion(
    model="openai/sonnet",
    api_base="http://localhost:52014/v1",
    api_key="not-needed",
    messages=[{"role": "user", "content": "Hello!"}]
)

Use with Vercel AI SDK

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { generateText } from "ai";

const claude = createOpenAICompatible({
  name: "claude-code-relay",
  baseURL: "http://localhost:52014/v1",
  apiKey: "not-needed",
});

const { text } = await generateText({
  model: claude.chatModel("sonnet"),
  prompt: "Hello!",
});

API Endpoints

Endpoint	Method	Description
/v1/chat/completions	POST	Chat completions (streaming supported)
/v1/models	GET	List available models
/health	GET	Health check

OpenAI API Compatibility

Supported Features

Feature	Status	Notes
model	Supported	sonnet, opus, haiku (+ aliases below)
messages	Supported	system, user, assistant roles
stream	Supported	SSE streaming
System prompts	Supported	Via system role in messages

Model Aliases

These model names are normalized to Claude CLI format:

Input	Maps to
sonnet	sonnet
opus	opus
haiku	haiku
claude-3-sonnet	sonnet
claude-3-opus	opus
claude-3-haiku	haiku
claude-sonnet-4	sonnet
claude-opus-4	opus

Not Supported

These parameters are accepted but ignored (not passed to Claude CLI):

Parameter	Status
temperature	Ignored
max_tokens	Ignored
top_p	Ignored
stop	Ignored
n	Not supported
presence_penalty	Not supported
frequency_penalty	Not supported
logit_bias	Not supported
response_format	Not supported
tools / functions	Not supported
tool_choice	Not supported
seed	Not supported
logprobs	Not supported
user	Not supported

Response Limitations

• usage tokens are always 0 (not tracked by Claude CLI)
• finish_reason is always "stop" (no length/tool_calls detection)

Configuration

Environment Variables

Variable	Default	Description
CLAUDE_CODE_RELAY_PORT	52014	Server port (Python only)
CLAUDE_CODE_RELAY_HOST	127.0.0.1	Host to bind (Python only)
CLAUDE_CLI_PATH	claude	Path to Claude CLI binary
CLAUDE_CODE_RELAY_TIMEOUT	300	Request timeout in seconds
CLAUDE_CODE_RELAY_VERBOSE	false	Enable verbose logging (1 or true)

CLI Options

claude-code-relay serve [options]

Options:
  --port, -p <port>      Server port (default: 52014)
  --host <host>          Host to bind (default: 127.0.0.1)
  --claude-path <path>   Path to Claude CLI
  --timeout <seconds>    Request timeout (default: 300)
  --verbose, -v          Enable verbose logging

Requirements

• Claude CLI installed and authenticated
• Python 3.10+ or Node.js 18+

License

MIT - see LICENSE

Disclaimer: Unofficial community project. Users are responsible for compliance with Anthropic's Terms of Service.