Ralph MCP Server

Ralph is a task-based AI orchestrator, now modernized as a Node.js Model Context Protocol (MCP) server.

Features

• Multi-Agent Support: Gemini (CLI), OpenAI (API), and Ollama (Local).
• Task Orchestration: Manages state, guardrails, and progress in .ralph directory.
• MCP Integration: Exposes Ralph logic as actionable tools for AI clients.

Installation

Via NPM (Official)

Once published, you can install Ralph MCP globally:

npm install -g @reaper1067msx/ralph-mcp

From Source

git clone https://github.com/craigm26/Ralph.git
cd Ralph
npm install
npm run build
npm link # For local development

Usage as MCP Server

Register in Gemini CLI

Add the following to your config:

{
  "mcpServers": {
    "ralph": {
      "command": "node",
      "args": ["C:/absolute/path/to/Ralph/dist/index.js"]
    }
  }
}

Tools Provided

• run_ralph_iteration(taskName, agentName?): Runs a single loop iteration for the specified task.
• get_ralph_status(): Returns the engine status and state directory.

Configuration

Create a ralph-config.json in your working directory:

{
  "agents": [
    {
      "name": "gemini-cli",
      "type": "gemini"
    },
    {
      "name": "my-gpt4",
      "type": "openai",
      "model": "gpt-4"
    }
  ],
  "defaultAgent": "gemini-cli",
  "stateDirectory": ".ralph"
}

Development

• npm run build: Compile TypeScript.
• npm test: Run unit and E2E tests.