Claude Code against an OpenAI-compatible backend

1. 01

   Understand the bridge requirement

   Claude Code calls POST /v1/messages with Anthropic-shaped bodies, which this gateway does not serve — it exposes the OpenAI chat-completions format. A local translation proxy is required; pointing ANTHROPIC_BASE_URL straight at the gateway will fail.

2. 02

   Install the router

   claude-code-router runs a local Anthropic-compatible server and forwards to any OpenAI provider.

   Install

   npm install -g @musistudio/claude-code-router

3. 03

   Register UncensoredAI as a provider

   Add the provider to the router config. api_base_url must include the full /v1/chat/completions path because the router appends nothing.

   ~/.claude-code-router/config.json

   {
     "Providers": [
       {
         "name": "uncensoredai",
         "api_base_url": "https://uncensoredai.dev/v1/chat/completions",
         "api_key": "sk-ua-XXXX-XXXX-XXXX-XXXX",
         "models": ["deepseek-v4.1-flash", "gpt-6-sol", "glm-5.3-flash-uncensored"]
       }
     ],
     "Router": {
       "default": "uncensoredai,deepseek-v4.1-flash",
       "background": "uncensoredai,deepseek-v4.1-flash",
       "think": "uncensoredai,gpt-6-sol"
     }
   }

4. 04

   Launch Claude Code through the router

   The router starts its local server and launches Claude Code against it. Every request bills from your UncensoredAI balance.

   Launch

   ccr code

Commands and configuration below are shown exactly as the tool expects them, so they stay in English.