local-openai2anthropic 0.1.0

A lightweight proxy server that converts Anthropic Messages API to OpenAI API

local-openai2anthropic

English | 中文

---

English

A lightweight proxy server that converts Anthropic Messages API requests to OpenAI API calls.

This allows you to use applications built for Claude's API with any OpenAI-compatible backend (OpenAI, Azure, local vLLM, etc.).

Features

• ✅ Full Messages API compatibility - All Anthropic Messages API features
• ✅ Streaming support - Server-sent events (SSE) for real-time responses
• ✅ Tool/Function calling - Convert between Anthropic tools and OpenAI functions
• ✅ Vision/Multimodal - Image input support
• ✅ Official SDK types - Uses official anthropic and openai Python SDKs for type safety
• ✅ Easy configuration - Environment variables or .env file
• ✅ CORS support - Ready for browser-based applications
• ✅ Self-hosted - Run locally or deploy to your infrastructure

Quick Start

Installation

# Install from source (recommended for now)
git clone https://github.com/yourusername/local-openai2anthropic.git
cd local-openai2anthropic
pip install -e ".[dev]"

# Or install directly
pip install local-openai2anthropic

Configuration

Set your OpenAI API key:

export OA2A_OPENAI_API_KEY="sk-..."

Or create a .env file:

OA2A_OPENAI_API_KEY=sk-...
OA2A_OPENAI_BASE_URL=https://api.openai.com/v1
OA2A_HOST=0.0.0.0
OA2A_PORT=8080

Run the Server

# Using the CLI command
local-openai2anthropic

# Or using the short alias
oa2a

# Or using Python module
python -m local_openai2anthropic

Usage Examples

Using with Anthropic Python SDK

import anthropic

# Point to your local proxy instead of Anthropic's API
client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",  # Not used but required by SDK
)

# Use the Messages API normally
message = client.messages.create(
    model="gpt-4o",  # This will be passed to OpenAI
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Hello, Claude!"}
    ]
)

print(message.content[0].text)

Using with Streaming

import anthropic

client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",
)

stream = client.messages.create(
    model="gpt-4o",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Count to 10"}],
    stream=True,
)

for event in stream:
    if event.type == "content_block_delta":
        print(event.delta.text, end="", flush=True)

Using with Tool Calling

import anthropic

client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",
)

message = client.messages.create(
    model="gpt-4o",
    max_tokens=1024,
    tools=[
        {
            "name": "get_weather",
            "description": "Get weather for a location",
            "input_schema": {
                "type": "object",
                "properties": {
                    "location": {"type": "string"},
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        }
    ],
    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}],
)

if message.stop_reason == "tool_use":
    tool_use = message.content[-1]
    print(f"Tool called: {tool_use.name}")
    print(f"Input: {tool_use.input}")

Using with Vision (Images)

import anthropic
import base64

client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",
)

# Read image and encode as base64
with open("image.png", "rb") as f:
    image_data = base64.b64encode(f.read()).decode()

message = client.messages.create(
    model="gpt-4o",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What's in this image?"},
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "image/png",
                        "data": image_data,
                    },
                },
            ],
        }
    ],
)

print(message.content[0].text)

Configuration Options

All configuration is done via environment variables (with OA2A_ prefix) or a .env file:

Variable	Default	Description
OA2A_OPENAI_API_KEY	Required	Your OpenAI API key
OA2A_OPENAI_BASE_URL	https://api.openai.com/v1	OpenAI API base URL
OA2A_OPENAI_ORG_ID	None	OpenAI organization ID
OA2A_OPENAI_PROJECT_ID	None	OpenAI project ID
OA2A_HOST	0.0.0.0	Server host to bind
OA2A_PORT	8080	Server port
OA2A_REQUEST_TIMEOUT	300.0	Request timeout in seconds
OA2A_API_KEY	None	Optional API key to protect your proxy
OA2A_CORS_ORIGINS	["*"]	Allowed CORS origins
OA2A_LOG_LEVEL	INFO	Logging level

Use Cases

1. Using Claude-based Apps with OpenAI Models

Many applications are built specifically for Claude's API. This proxy lets you use them with GPT-4, GPT-3.5, or any OpenAI-compatible model.

2. Local Development with vLLM

Run a local vLLM server with OpenAI-compatible API, then use this proxy to test Claude-integrated apps:

# Terminal 1: Start vLLM
vllm serve meta-llama/Llama-2-7b-chat-hf --api-key dummy

# Terminal 2: Start proxy pointing to vLLM
export OA2A_OPENAI_API_KEY=dummy
export OA2A_OPENAI_BASE_URL=http://localhost:8000/v1
local-openai2anthropic

# Terminal 3: Use Claude SDK with local model
python my_claude_app.py  # Point to http://localhost:8080

3. Azure OpenAI Service

export OA2A_OPENAI_API_KEY="your-azure-key"
export OA2A_OPENAI_BASE_URL="https://your-resource.openai.azure.com/openai/deployments/your-deployment"
local-openai2anthropic

4. Other OpenAI-Compatible APIs

• Groq
• Together AI
• Fireworks
• Anyscale
• LocalAI
• llama.cpp server

API Coverage

Supported Anthropic Features

Feature	Status	Notes
messages.create()	✅ Full	All parameters supported
Streaming	✅ Full	SSE with all event types
Tool use	✅ Full	Converted to OpenAI functions
Vision	✅ Full	Images converted to base64 data URLs
System prompts	✅ Full	String or array format
Stop sequences	✅ Full	Passed through
Temperature	✅ Full
Top P	✅ Full
Top K	✅ Full
Max tokens	✅ Full
Thinking	⚠️ Partial	Mapped to reasoning_effort where supported

Not Supported

• Prompt caching (cache_control) - OpenAI doesn't have equivalent
• Computer use (beta) - Requires native Claude capabilities

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    Your Application                             │
│              (uses Anthropic Python SDK)                        │
└───────────────────────┬─────────────────────────────────────────┘
                        │ Anthropic Messages API format
                        ▼
┌─────────────────────────────────────────────────────────────────┐
│              local-openai2anthropic Proxy                       │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐      │
│  │   Anthropic  │───▶│   Converter  │───▶│    OpenAI    │      │
│  │   Request    │    │              │    │   Request    │      │
│  └──────────────┘    └──────────────┘    └──────────────┘      │
│                                                               │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐      │
│  │   Anthropic  │◀───│   Converter  │◀───│    OpenAI    │      │
│  │   Response   │    │              │    │   Response   │      │
│  └──────────────┘    └──────────────┘    └──────────────┘      │
└───────────────────────┬─────────────────────────────────────────┘
                        │ OpenAI API format
                        ▼
┌─────────────────────────────────────────────────────────────────┐
│              OpenAI-compatible Backend                          │
│        (OpenAI, Azure, vLLM, Groq, etc.)                        │
└─────────────────────────────────────────────────────────────────┘

Development

# Clone repository
git clone https://github.com/yourusername/local-openai2anthropic.git
cd local-openai2anthropic

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black src/
ruff check src/

# Type check
mypy src/

License

Apache License 2.0 - See LICENSE for details.

Acknowledgments

This project is based on the Anthropic API implementation from vLLM, adapted to work as a standalone proxy service.

---

中文

一个轻量级代理服务器，将 Anthropic Messages API 请求转换为 OpenAI API 调用。

让你可以使用任何 OpenAI 兼容的后端（OpenAI、Azure、本地 vLLM 等）来运行为 Claude API 构建的应用程序。

功能特性

• ✅ 完整的 Messages API 兼容性 - 支持所有 Anthropic Messages API 功能
• ✅ 流式响应支持 - 使用服务器推送事件（SSE）实现实时响应
• ✅ 工具/函数调用 - Anthropic 工具与 OpenAI 函数之间的双向转换
• ✅ 视觉/多模态 - 支持图片输入
• ✅ 官方 SDK 类型 - 使用官方 anthropic 和 openai Python SDK 确保类型安全
• ✅ 简单配置 - 支持环境变量或 .env 文件
• ✅ CORS 支持 - 开箱即用的浏览器应用支持
• ✅ 自托管 - 可在本地运行或部署到自己的基础设施

快速开始

安装

# 从源码安装（当前推荐）
git clone https://github.com/yourusername/local-openai2anthropic.git
cd local-openai2anthropic
pip install -e ".[dev]"

# 或直接安装
pip install local-openai2anthropic

配置

设置你的 OpenAI API 密钥：

export OA2A_OPENAI_API_KEY="sk-..."

或创建 .env 文件：

OA2A_OPENAI_API_KEY=sk-...
OA2A_OPENAI_BASE_URL=https://api.openai.com/v1
OA2A_HOST=0.0.0.0
OA2A_PORT=8080

运行服务器

# 使用 CLI 命令
local-openai2anthropic

# 或使用短别名
oa2a

# 或使用 Python 模块
python -m local_openai2anthropic

使用示例

使用 Anthropic Python SDK

import anthropic

# 指向本地代理而非 Anthropic 官方 API
client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",  # 需要填写但不会被使用
)

# 像平常一样使用 Messages API
message = client.messages.create(
    model="gpt-4o",  # 会传递给 OpenAI
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "你好，Claude！"}
    ]
)

print(message.content[0].text)

使用流式响应

import anthropic

client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",
)

stream = client.messages.create(
    model="gpt-4o",
    max_tokens=1024,
    messages=[{"role": "user", "content": "数到 10"}],
    stream=True,
)

for event in stream:
    if event.type == "content_block_delta":
        print(event.delta.text, end="", flush=True)

使用工具调用

import anthropic

client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",
)

message = client.messages.create(
    model="gpt-4o",
    max_tokens=1024,
    tools=[
        {
            "name": "get_weather",
            "description": "获取某地天气",
            "input_schema": {
                "type": "object",
                "properties": {
                    "location": {"type": "string"},
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        }
    ],
    messages=[{"role": "user", "content": "东京天气如何？"}],
)

if message.stop_reason == "tool_use":
    tool_use = message.content[-1]
    print(f"调用工具: {tool_use.name}")
    print(f"输入参数: {tool_use.input}")

使用视觉（图片）

import anthropic
import base64

client = anthropic.Anthropic(
    base_url="http://localhost:8080",
    api_key="dummy-key",
)

# 读取图片并编码为 base64
with open("image.png", "rb") as f:
    image_data = base64.b64encode(f.read()).decode()

message = client.messages.create(
    model="gpt-4o",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "这张图片里有什么？"},
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "image/png",
                        "data": image_data,
                    },
                },
            ],
        }
    ],
)

print(message.content[0].text)

配置选项

所有配置都可以通过环境变量（前缀为 OA2A_）或 .env 文件设置：

变量	默认值	说明
OA2A_OPENAI_API_KEY	必填	你的 OpenAI API 密钥
OA2A_OPENAI_BASE_URL	https://api.openai.com/v1	OpenAI API 基础地址
OA2A_OPENAI_ORG_ID	None	OpenAI 组织 ID
OA2A_OPENAI_PROJECT_ID	None	OpenAI 项目 ID
OA2A_HOST	0.0.0.0	服务器绑定的主机地址
OA2A_PORT	8080	服务器端口
OA2A_REQUEST_TIMEOUT	300.0	请求超时时间（秒）
OA2A_API_KEY	None	保护代理的可选 API 密钥
OA2A_CORS_ORIGINS	["*"]	允许的 CORS 来源
OA2A_LOG_LEVEL	INFO	日志级别

使用场景

1. 使用 Claude 应用配合 OpenAI 模型

许多应用是专门为 Claude API 构建的。此代理允许你将它们与 GPT-4、GPT-3.5 或任何 OpenAI 兼容模型一起使用。

2. 配合 vLLM 进行本地开发

运行本地 vLLM 服务器，使用 OpenAI 兼容 API，然后通过此代理测试集成 Claude 的应用：

# 终端 1：启动 vLLM
vllm serve meta-llama/Llama-2-7b-chat-hf --api-key dummy

# 终端 2：启动代理指向 vLLM
export OA2A_OPENAI_API_KEY=dummy
export OA2A_OPENAI_BASE_URL=http://localhost:8000/v1
local-openai2anthropic

# 终端 3：使用 Claude SDK 配合本地模型
python my_claude_app.py  # 指向 http://localhost:8080

3. Azure OpenAI 服务

export OA2A_OPENAI_API_KEY="your-azure-key"
export OA2A_OPENAI_BASE_URL="https://your-resource.openai.azure.com/openai/deployments/your-deployment"
local-openai2anthropic

4. 其他 OpenAI 兼容 API

• Groq
• Together AI
• Fireworks
• Anyscale
• LocalAI
• llama.cpp server

API 覆盖情况

支持的 Anthropic 功能

功能	状态	备注
messages.create()	✅ 完整	支持所有参数
流式响应	✅ 完整	SSE 支持所有事件类型
工具调用	✅ 完整	转换为 OpenAI 函数
视觉	✅ 完整	图片转换为 base64 data URLs
系统提示词	✅ 完整	支持字符串或数组格式
停止序列	✅ 完整	透传
温度	✅ 完整
Top P	✅ 完整
Top K	✅ 完整
最大 token	✅ 完整
思考模式	⚠️ 部分	在支持的模型上映射到 reasoning_effort

不支持的功能

• 提示词缓存 (cache_control) - OpenAI 没有等效功能
• 计算机使用（测试版） - 需要原生 Claude 能力

架构

┌─────────────────────────────────────────────────────────────────┐
│                    你的应用                                       │
│              (使用 Anthropic Python SDK)                          │
└───────────────────────┬─────────────────────────────────────────┘
                        │ Anthropic Messages API 格式
                        ▼
┌─────────────────────────────────────────────────────────────────┐
│              local-openai2anthropic 代理                          │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐      │
│  │   Anthropic  │───▶│   Converter  │───▶│    OpenAI    │      │
│  │   Request    │    │              │    │   Request    │      │
│  └──────────────┘    └──────────────┘    └──────────────┘      │
│                                                               │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐      │
│  │   Anthropic  │◀───│   Converter  │◀───│    OpenAI    │      │
│  │   Response   │    │              │    │   Response   │      │
│  └──────────────┘    └──────────────┘    └──────────────┘      │
└───────────────────────┬─────────────────────────────────────────┘
                        │ OpenAI API 格式
                        ▼
┌─────────────────────────────────────────────────────────────────┐
│              OpenAI 兼容后端                                      │
│        (OpenAI、Azure、vLLM、Groq 等)                            │
└─────────────────────────────────────────────────────────────────┘

开发

# 克隆仓库
git clone https://github.com/yourusername/local-openai2anthropic.git
cd local-openai2anthropic

# 以开发模式安装
pip install -e ".[dev]"

# 运行测试
pytest

# 格式化代码
black src/
ruff check src/

# 类型检查
mypy src/

许可证

Apache License 2.0 - 详情请参阅 LICENSE。

致谢

本项目基于 vLLM 的 Anthropic API 实现，适配为独立的代理服务。