This repository lets you use Anthropic's Claude Code CLI with OpenAI models such as GPT-5-Codex, GPT-5.1, and others via a local LiteLLM proxy.

⚠️ ATTENTION ⚠️

If you're here to set up your own LiteLLM Server with LibreChat as the web UI (or any other OpenAI / Anthropic API compatible client, for that matter), head over to the litellm-server-boilerplate repository. It contains a "boilerplate" version of this repo with Claude Code CLI stuff stripped away, an optional LibreChat set up, and a README which specifically explains how to build your own AI agents and assistants on top of it.

Quick Start ⚡

Prerequisites

• OpenAI API key 🔑
• Anthropic API key 🔑 - optional (if you decide not to remap some Claude models to OpenAI)
• Either uv or Docker Desktop, depending on your preferred setup method

First time using GPT-5 via API?

If you are going to use GPT-5 via API for the first time, OpenAI may require you to verify your identity via Persona. You may encounter an OpenAI error asking you to “verify your organization.” To resolve this, you can go through the verification process here:

• OpenAI developer platform - Organization settings

Setup 🛠️

1. Clone this repository:

   git clone https://github.com/teremterem/claude-code-gpt-5-codex.git
   cd claude-code-gpt-5-codex

2. Configure Environment Variables:

   Copy the template file to create your .env:

   cp .env.template .env

   Edit .env and add your OpenAI API key:

   OPENAI_API_KEY=your-openai-api-key-here
   # Optional: only needed if you plan to use Anthropic models
   # ANTHROPIC_API_KEY=your-anthropic-api-key-here
   
   # Optional (see .env.template for details):
   # LITELLM_MASTER_KEY=your-master-key-here
   
   # Optional: specify the remaps explicitly if you need to (the values you see
   # below are the defaults - see .env.template for more info)
   # REMAP_CLAUDE_HAIKU_TO=gpt-5.1-codex-mini-reason-none
   # REMAP_CLAUDE_SONNET_TO=gpt-5-codex-reason-medium
   # REMAP_CLAUDE_OPUS_TO=gpt-5.1-reason-high
   
   # Some more optional settings (see .env.template for details)
   ...

3. Run the proxy:

   1. EITHER via uv (make sure to install uv first):

      OPTION 1: Use a script for uv:

      ./uv-run.sh

      OPTION 2: Run via a direct uv command:

      uv run litellm --config config.yaml

   2. OR via Docker (make sure to install Docker Desktop first):

      OPTION 3: Run Docker in the foreground:

      ./run-docker.sh

      OPTION 4: Run Docker in the background:

      ./deploy-docker.sh

      OPTION 5: Run Docker via a direct command:

      docker run -d \
         --name claude-code-gpt-5 \
         -p 4000:4000 \
         --env-file .env \
         --restart unless-stopped \
         ghcr.io/teremterem/claude-code-gpt-5:latest

      NOTE: To run with this command in the foreground instead of the background, remove the -d flag.

      To see the logs, run:

      docker logs -f claude-code-gpt-5

      To stop and remove the container, run:

      ./kill-docker.sh

      NOTE: The Docker options above will pull the latest image from GHCR and will ignore all your local files except .env. For more detailed Docker deployment instructions and more options (like building Docker image from source yourself, using Docker Compose, etc.), see docs/DOCKER_TIPS.md

Using with Claude Code 🎮

1. Install Claude Code (if you haven't already):

   npm install -g @anthropic-ai/claude-code

2. Connect it to the proxy:

   ANTHROPIC_BASE_URL=http://localhost:4000 claude

   If you set LITELLM_MASTER_KEY in your .env file (see .env.template for details), pass it as the Anthropic API key for the CLI:

   ANTHROPIC_API_KEY="<LITELLM_MASTER_KEY>" \
   ANTHROPIC_BASE_URL=http://localhost:4000 \
   claude

   NOTE: In this case, if you've previously authenticated, run claude /logout first.

3. That's it! Your Claude Code client will now use the OpenAI models that this repo recommends by default (unless you explicitly specified different choices in your .env file). 🎯

Model aliases

You can find the full list of available OpenAI models in the OpenAI API documentation. Additionally, this proxy allows you to control the reasoning effort level for each model by appending it to the model name following the pattern -reason-<effort> (or -reasoning-<effort>, if you prefer). Here are some examples:

• gpt-5.1-codex-mini-reason-none
• gpt-5.1-codex-mini-reason-medium
• gpt-5.1-codex-mini-reason-high

If you don't specify the reasoning effort level (i.e. only specify the model name, like gpt-5.1-codex-mini), it will use the default level for the model.

NOTE: Theoretically, you can use arbitrary models from arbitrary providers, but for providers other than OpenAI or Anthropic, you will need to specify the provider as a prefix in the model name, e.g. gemini/gemini-pro, gemini/gemini-pro-reason-disable etc. (as well as set the respective API key for that provider in your .env file).

KNOWN PROBLEM

The Web Search tool currently does not work with this setup. You may see an error like:

API Error (500 {"error":{"message":"Error calling litellm.acompletion for non-Anthropic model: litellm.BadRequestError: OpenAIException - Invalid schema for function 'web_search': 'web_search_20250305' is not valid under any of the given schemas.","type":"None","param":"None","code":"500"}}) · Retrying in 1 seconds… (attempt 1/10)

This is planned to be fixed soon.

NOTE: The Fetch tool (getting web content from specific URLs) is not affected and works normally.

P. S. You are welcome to join our MiniAgents Discord Server 👥

And if you like the project, please give it a Star 💫