A proxy service that allows Anthropic API requests (especially from Claude Code to be routed through an OpenAI-compatible URL to access alternative models.

• Overview
• Example
• Getting Started
  • Prerequisites
• Misc
• Known Working Models
• License

Claude Proxy provides a compatibility layer between Claude Code and alternative models available through OpenRouter.ai or your chosen base URL. It dynamically reroutes LLM requests from Claude Code to the providers you want to use.

Key features:

• FastAPI web server exposing Anthropic-compatible endpoints
• Format conversion between Anthropic and OpenAI requests/responses (see mapping for translation details)
• Support for both streaming and non-streaming responses
• Dynamic model selection based on requested Claude model
• Detailed request/response logging
• Token counting

It is recommended to use docker, using uv is recommended for development only.

• Claude Code installed (as of July 2025 the proxy is tested on version 1.0.56, which you can install with npm install -g @anthropic-ai/[email protected])
• An OpenRouter API key
• Optional if you don't want to use docker
  • Python 3.10+
  • uv

1. Download the repo: git clone https://github.com/ujisati/claude-code-provider-proxy/

2. Either modify the environment: in docker-compose or create a .env file at the root of the repo with for example:

OPENAI_API_KEY=<your openrouter api key>
BIG_MODEL_NAME=google/gemini-2.5-pro-preview
SMALL_MODEL_NAME=google/gemini-2.0-flash-lite-001
LOG_LEVEL=DEBUG

A list of known working models can be found at the bottom of that README.md.

For more configuration options, see the Settings class in src/main.py.

Note: you can use environment variables instead of an .env file, but note that any value of the .env file that is also present in the environment will be overwritten (the environment takes priority). In particular, if you set OPENAI_API_KEY in .env to the OPENROUTER_API_KEY, if you don't unset OPENAI_API_KEY, the key received by src/main.py will be the one from OpenAI instead of OpenRouter.ai.

3. Run the proxy:

Docker: (recommended)

# Build and run with docker-compose
docker-compose up --build --detach

# Or build and run manually
docker build -f docker/Dockerfile -t claude-code-proxy .
docker run -p 8080:8080 --env-file .env claude-code-proxy

Local Development:

uv run src/main.py

4. Run Claude Code

ANTHROPIC_BASE_URL=http://localhost:8080 claude

To make a more permanent alias you can run echo 'alias claude="ANTHROPIC_BASE_URL=http://localhost:8080 claude"' >> ~/.bashrc for example.

5. Optional: to further customize claude, there are environment variables you can modify, for example:

CLAUDE_CODE_EXTRA_BODY MAX_THINKING_TOKENS API_TIMEOUT_MS ANTHROPIC_BASE_URL DISABLE_TELEMETRY DISABLE_ERROR_REPORTING

Note: These variables modify claude, not the proxy.

The proxy server exposes the following endpoints:

• POST /v1/messages: Create a message (main endpoint)
• POST /v1/messages/count_tokens: Count tokens for a request
• GET /: Health check endpoint

• As BIG_MODEL_NAME:

  • anthropic/claude-sonnet-4

  • problematic

    • google/gemini-2.5-pro: seems to sometimes struggle to respect the edit format expected by claude

• As SMALL_MODEL_NAME:

  • anthropic/claude-3.5-haiku
  • google/gemini-2.5-flash

LICENSE