MClaude v1.8.6

About

MClaude allows you to pick from any model within the setup for the 4 models that Claude uses. Currently supports Synthetic.New and MiniMax with more networks being added regularly.

🚀 Quick Start

npm install -g mclaude
mclaude setup
mclaude models  # Interactive model selection

🎯 Key Features

• Zero Dependencies Proxy: Self-contained HTTP server (port 9313), no Docker/pip needed
• Smart Model Routing: minimax:*, synthetic:*, auto-prefix fuzzy matching
• Dynamic Model Discovery: Real-time model lists from provider APIs
• Full Anthropic API: /v1/messages?beta=true endpoint compatibility
• CLI-First: mclaude --model minimax:MiniMax-M2 launches instantly

📋 Commands

mclaude models                    # Interactive model picker
mclaude --model claude-3-sonnet   # Direct launch (auto-detects provider)
mclaude doctor                    # System health check
mclaude setup                     # Initial setup + proxy install
mclaude providers                 # Manage API keys
mclaude cache clear               # Clear model cache

🛠️ Architecture

Claude Code → Custom Proxy (localhost:9313) → MiniMax/Synthetic AI

⚙️ Configuration

~/.config/mclaude/config.json
# or .mclaude/config.json (project-local)

🔧 Development

npm install
npm run dev      # ts-node src/index.ts
npm test         # 133/133 tests
npm run build    # tsc + chmod +x
npm version patch # Triggers GitHub Actions publish

📦 Installation Methods

1. npm (Recommended)

   npm install -g mclaude

2. npx

   npx mclaude@latest setup

3. Shell Script

   curl -fsSL https://mclaude.sh | bash

CLI Commands

Proxy Management

# Start the TensorZero proxy (auto-detects if already running)
mclaude proxy start --verbose

# Check proxy status (shows URL, available routes, uptime)
mclaude proxy status

# Stop the proxy (properly kills all Python processes)
mclaude proxy stop --verbose

# Restart the proxy (full stop/start cycle)
mclaude proxy restart

Model Selection

# Interactive model selection
mclaude models

# Direct model launch
mclaude --model synthetic:deepseek-ai/DeepSeek-V3.2
mclaude --model minimax:MiniMax-M2 --dangerously-skip-permissions

# Quick test command
echo "Hello, Claude!" | mclaude --model synthetic:meta-llama/Llama-3.2-3B-Instruct

Configuration & Management

# Initial setup
mclaude setup

# System health check
mclaude doctor

# Provider management
mclaude providers
mclaude auth show synthetic

Proxy Management

The TensorZero proxy provides model routing on port 9313. Only one proxy instance runs per system - CLI commands automatically detect existing instances.

Key Features:

• Collision Detection: Won't start duplicate proxies
• Process Cleanup: Actually kills detached Python processes
• Cross-Instance Status: Status detection works between different CLI invocations
• Auto-Configuration: Enables tensorzero.enabled: true in config automatically
• 26 Available Models: Synthetic and MiniMax models routed through /v1/models

Model Naming Patterns:

• ✅ synthetic:deepseek-ai/DeepSeek-V3.2
• ✅ synthetic:meta-llama/Llama-3.2-3B-Instruct
• ✅ minimax:MiniMax-M2
• ❌ hf:deepseek-ai/DeepSeek-V3.2 (proxies use synthetic: prefix)

Previous Versions

See CHANGELOG.md for full history