MyAgent - AI Agent Framework with RAG and MCP Integration

A sophisticated AI agent framework built with TypeScript that combines Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and Model Context Protocol (MCP) for enhanced AI capabilities.

🚀 Features

• 🤖 AI Agent: Autonomous agent with tool-calling capabilities
• 🔍 RAG System: Retrieval-Augmented Generation with embedding-based search
• 🛠️ MCP Integration: Model Context Protocol for external tool integration
• 📊 Vector Store: In-memory vector database with cosine similarity search
• 🌊 Streaming Support: Real-time streaming responses from OpenAI
• 🔧 Multiple Tools: File system operations, web fetching, and more
• 🎯 TypeScript: Full type safety and modern development experience

🏗️ Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│      Agent      │───▶│    ChatOpenAI    │───▶│   OpenAI API    │
│  (Coordinator)  │    │  (LLM Interface) │    │                 │
└─────────────────┘    └──────────────────┘    └─────────────────┘
         │
         ▼
┌─────────────────┐    ┌──────────────────┐
│    MCPClient    │───▶│   Tool Servers   │
│ (Tool Interface)│    │  (File, Web...)  │
└─────────────────┘    └──────────────────┘
         │
         ▼
┌──────────────────┐    ┌──────────────────┐
│EmbeddingRetriever│───▶│   Vector Store   │
│   (RAG System)   │    │   (Similarity)   │
└──────────────────┘    └──────────────────┘

📁 Project Structure

src/
├── Agent.ts              # Main agent coordinator
├── ChatOpenAI.ts         # OpenAI API wrapper with streaming
├── MCPClient.ts          # Model Context Protocol client
├── EmbeddingRetriever.ts # RAG embedding and retrieval
├── VectorStore.ts        # In-memory vector database
├── utils.ts              # Utility functions
└── index.ts              # Example implementation

🛠️ Installation

# Clone the repository
git clone <repository-url>
cd myagent

# Install dependencies
pnpm install

# Set up environment variables
cp .env.example .env

⚙️ Configuration

Create a .env file with the following variables:

# OpenAI Configuration
OPENAI_API_KEY=your_openai_api_key
OPENAI_API_BASE_URL=https://api.openai.com/v1

# Embedding Service Configuration
EMBEDDING_BASE_URL=your_embedding_service_url
EMBEDDING_KEY=your_embedding_api_key

# Proxy Configuration (optional)
HTTPS_PROXY=http://127.0.0.1:7890

🚀 Quick Start

Basic Agent Usage

import Agent from "./Agent";
import MCPClient from "./MCPClient";

// Create MCP clients for different tools
const fileMCP = new MCPClient("file-server", "npx", ['-y', '@modelcontextprotocol/server-filesystem', './workspace']);
const fetchMCP = new MCPClient("fetch-server", "uvx", ['mcp-server-fetch']);

// Initialize agent
const agent = new Agent('gpt-4o-mini', [fileMCP, fetchMCP]);
await agent.init();

// Use the agent
const response = await agent.invoke("Please read the README.md file and summarize it");
console.log(response);

// Clean up
await agent.close();

RAG System Usage

import EmbeddingRetriever from "./EmbeddingRetriever";

// Initialize retriever
const retriever = new EmbeddingRetriever("BAAI/bge-m3");

// Add documents
await retriever.embedDocument("Your document content here");

// Retrieve relevant content
const results = await retriever.retrieve("Your query", 3);
console.log(results);

Streaming Chat

import ChatOpenAI from "./ChatOpenAI";

const llm = new ChatOpenAI("gpt-4o-mini", "You are a helpful assistant");
const { content, toolCalls } = await llm.chat("Hello, how are you?");
console.log(content); // Streams in real-time

🧩 Core Components

Agent Class

The main coordinator that orchestrates interactions between LLMs and tools.

Key Features:

• Autonomous tool calling
• Multi-round conversations
• Automatic tool result handling
• Resource management

ChatOpenAI Class

OpenAI API wrapper with streaming support and tool integration.

Key Features:

• Real-time streaming responses
• Tool call processing
• Conversation history management
• Proxy support for restricted networks

MCPClient Class

Model Context Protocol client for external tool integration.

Key Features:

• Stdio transport for tool servers
• Dynamic tool discovery
• Tool execution management
• Connection lifecycle handling

EmbeddingRetriever Class

RAG system with embedding-based document retrieval.

Key Features:

• Document embedding
• Similarity search
• Top-k retrieval
• Multiple embedding model support

VectorStore Class

In-memory vector database with similarity search.

Key Features:

• Cosine similarity calculation
• Efficient vector storage
• Top-k search results
• Simple API interface

📝 Scripts

# Development
pnpm dev          # Run in development mode

# Production
pnpm build        # Compile TypeScript
pnpm start        # Run compiled JavaScript

# Example Usage
pnpm dev          # Runs the example in src/index.ts

🔧 Available MCP Servers

The project supports various MCP servers for different functionalities:

• File System: @modelcontextprotocol/server-filesystem
• Web Fetching: mcp-server-fetch
• Database: Various database connectors
• Custom Tools: Easy to integrate custom MCP servers

🌟 Example Use Cases

1. Document Analysis: RAG-powered document Q&A system
2. Web Research: Autonomous web scraping and analysis
3. File Management: AI-powered file operations
4. Content Generation: Context-aware content creation
5. Data Processing: Automated data analysis workflows

🔒 Security Notes

• API keys are loaded from environment variables
• Proxy support for restricted network environments
• Tool execution is sandboxed through MCP protocol
• No sensitive data is logged

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

📄 License

ISC License - see LICENSE file for details

🙏 Acknowledgments

• OpenAI for the GPT models
• Model Context Protocol for tool integration
• TypeScript for type safety
• Various MCP server implementations