Claude Code Toolkit

Connect Claude Code to self-hosted LLM services (vLLM, SGLang, or any OpenAI-compatible endpoint) for offline development and testing model agent capabilities.

Powered by claude-code-router — model routing with enhancetool for tool_use error tolerance.

Architecture

Your terminal
  │
  ▼
docker exec -it claude_container claude
  │
  ▼
┌─────────────────────────────────┐
│         Docker Container        │
│                                 │
│  Claude Code CLI                │
│       │                         │
│  claude-code-router (:3456)     │
│    (enhancetool + routing)      │
└───────┼─────────────────────────┘
        │
        ▼
  LLM Provider (OpenAI protocol)
  vLLM / SGLang / ...

With debug proxy (DEBUG=1):

Claude Code → debug-proxy (:8083) → ccr (:3456) → debug-proxy (:8084) → LLM Provider
               [cc2ccr.log]                         [ccr2provider.log]

claude-container

One Docker image with Claude Code + claude-code-router, connecting to any OpenAI-compatible endpoint.

See claude-container/README.md | 中文文档