Framework-RAI MCP Server

A Model Context Protocol (MCP) server for responsible AI development and documentation.

Features

• Project Scanning: Automatically detect AI libraries, model files, and training scripts in your project
• Responsible AI Suggestions: Generate suggestions for improving bias mitigation, transparency, privacy, and monitoring
• Model Analysis: Analyze model code for potential issues related to bias, documentation, security, and testing
• Documentation Management: Create and update responsible AI documentation including checklists and model cards

Installation

Using Smithery (Recommended)

If you're using Cursor or another Smithery-compatible client, you can install this package directly:

npx -y @smithery/cli@latest inspect @sebastianbuzdugan/framework-rai-mcp

When prompted by Smithery, you'll need to provide your OpenAI API key to use the AI-powered features.

Manual Installation

You can also install the package globally:

npm install -g framework-rai-mcp

Or install locally for development:

git clone https://github.com/sebastianbuzdugan/framework-rai-mcp.git
cd framework-rai-mcp
npm install

Usage

As a Smithery Plugin

Once installed via Smithery, the Framework-RAI functions will be available directly in your Smithery-compatible client like Cursor. The first time you use a function that requires OpenAI, you'll be prompted to enter your API key.

As a Command Line Tool

Start the MCP server:

framework-rai-mcp

By default, the server runs on port 3001. You can specify a different port:

framework-rai-mcp --port=3003

Setting Your OpenAI API Key

The AI-powered features require an OpenAI API key. You can provide it in several ways:

1. Command line argument:

   framework-rai-mcp --api-key=sk-your-openai-key

2. Environment variable:

   export OPENAI_API_KEY=sk-your-openai-key
   framework-rai-mcp

3. Create a .env file in your project directory:

   OPENAI_API_KEY=sk-your-openai-key

JSON-RPC API

The server implements the Model Context Protocol (MCP) using JSON-RPC 2.0 at the /mcp endpoint. The following methods are available:

• initialize: Initialize a session with the server
• tools/list: List available tools
• tools/call: Call a specific tool with parameters
• shutdown: Terminate a session

Available Tools

• scanProject: Scan a project for AI components
• generateSuggestions: Generate responsible AI suggestions (requires OpenAI API key)
• analyzeModel: Analyze a model file for potential issues (requires OpenAI API key)
• getDocumentation: Get responsible AI documentation
• updateDocumentation: Update responsible AI documentation

Testing the Server

You can test the server using the included test script:

npm run mcp-test

This will send test requests to the server and display the responses.

Environment Variables

• PORT: Port number for the server (default: 3001)
• OPENAI_API_KEY: Your OpenAI API key (required for suggestions and analysis)

Requirements

• Node.js 14 or higher
• OpenAI API key (for AI-powered features)

License

MIT

Author

Sebastian Buzdugan