Tetrad

Quadruple Consensus MCP Server for Claude Code

Tetrad is a high-performance MCP (Model Context Protocol) server that orchestrates three AI-powered CLI tools (Codex, Gemini CLI, Qwen) to validate all code produced by Claude Code.

Quick Start

1. Initialize in your project

npx @samoradc/tetrad init

This will:

• Create tetrad.toml configuration file
• Create .tetrad/ directory for the database
• Add .tetrad/ to your .gitignore

2. Add to Claude Code

# Add as MCP server (available in all projects)
claude mcp add --scope user tetrad -- npx @samoradc/tetrad serve

# Or for current project only
claude mcp add tetrad -- npx @samoradc/tetrad serve

3. Verify

# Check version
npx @samoradc/tetrad version

# Check CLI availability
npx @samoradc/tetrad status

# Diagnose issues
npx @samoradc/tetrad doctor

Commands

npx @samoradc/tetrad init              # Initialize config in current directory
npx @samoradc/tetrad serve             # Start MCP server (used by Claude Code)
npx @samoradc/tetrad status            # Show CLI status (codex, gemini, qwen)
npx @samoradc/tetrad config            # Interactive configuration
npx @samoradc/tetrad doctor            # Diagnose configuration issues
npx @samoradc/tetrad version           # Show version
npx @samoradc/tetrad evaluate -c CODE  # Manual code evaluation (without MCP)
npx @samoradc/tetrad history           # Show evaluation history from ReasoningBank

Manual MCP Configuration

Add to your .mcp.json:

{
  "mcpServers": {
    "tetrad": {
      "type": "stdio",
      "command": "npx",
      "args": ["@samoradc/tetrad", "serve"]
    }
  }
}

Requirements

Tetrad requires at least one of these AI CLI tools:

• Codex CLI (OpenAI): npm install -g @openai/codex
• Gemini CLI (Google): npm install -g @google/gemini-cli
• Qwen CLI (Alibaba): pip install dashscope

Features

• Quadruple Consensus: 4 AI models must agree to approve code
• ReasoningBank: Continuous learning system with SQLite
• High Performance: Written in Rust
• LRU Cache: Result caching with configurable TTL
• Hook System: Pre/post evaluation callbacks
• Auto .gitignore: Automatically ignores local data

Configuration

After npx @samoradc/tetrad init, edit tetrad.toml:

[general]
log_level = "info"
timeout_secs = 60

[executors.codex]
enabled = true
command = "codex"
args = ["exec", "--json"]

[executors.gemini]
enabled = true
command = "gemini"
args = ["-o", "json"]

[executors.qwen]
enabled = true
command = "qwen"

[consensus]
default_rule = "strong"  # golden, strong, weak
min_score = 70
max_loops = 3

[reasoning]
enabled = true
db_path = ".tetrad/tetrad.db"

[cache]
enabled = true
capacity = 1000
ttl_secs = 300

Documentation

Full documentation: GitHub Repository

License

MIT