이 저장소는 MCP (Model Context Protocol)과 통합된 Amazon Nova Act을 사용하여 지능형 웹 자동화 에이전트를 구축하는 방법을 보여줍니다. MCP는 다른 데이터 소스 및 도구에 AI 모델을 연결하는 표준화 된 방법을 제공합니다 - AI 애플리케이션의 "USB-C 포트"와 같은 생각.

개요

이 프로젝트는 다음과 같은 통합을 보여줍니다.

• Amazon Nova Act: 웹 페이지와 상호 작용하는 에이전트를 만드는 강력한 웹 자동화 도구
• MCP (모델 컨텍스트 프로토콜): 애플리케이션이 LLMs에 대한 컨텍스트를 제공하는 방법을 표준화하는 개방형 프로토콜
• **Amazon Bedrock **: 통합된 API를 통해 기초 모델을 제공하는 관리 서비스

건축

프로젝트는 MCP의 클라이언트 서버 아키텍처를 따릅니다:

• MCP Host: Nova Act 기능에 액세스하려면 Claude Desktop 또는 기타 AI 도구
• MCP Client: Nova Act 서버와 1:1 연결을 유지
• MCP Server: 표준화된 프로토콜을 통해 Nova Act 서버가 웹 자동화 기능을 폭발
• Remote Services: Nova Act가 브라우저 자동화를 통해 상호 작용하는 웹 서비스

키 기능

• MCP를 통한 표준화된 LLM 통합

Building Agents with Amazon Nova Act and MCP

This repository demonstrates how to build intelligent web automation agents using Amazon Nova Act integrated with MCP (Model Context Protocol). MCP provides a standardized way to connect AI models to different data sources and tools - think of it like a “USB-C port for AI applications.”

Overview

This project showcases the integration between:

• Amazon Nova Act: A powerful web automation tool for creating agents that interact with web pages
• MCP (Model Context Protocol): An open protocol that standardizes how applications provide context to LLMs
• Amazon Bedrock: Managed service that provides foundation models through a unified API

Architecture

The project follows MCP’s client-server architecture:

• MCP Host: Claude Desktop or other AI tools that want to access Nova Act capabilities
• MCP Client: Protocol clients maintaining 1:1 connections with the Nova Act server
• MCP Server: A Nova Act server exposing web automation capabilities through the standardized protocol
• Remote Services: Web services that Nova Act interacts with through browser automation

Key Features

• Standardized LLM integration through MCP
• Automated web navigation and interaction via Nova Act
• Parallel execution of multiple browser tasks
• Data extraction and processing
• Screenshot capture and visual verification
• Session management and browser control
• Seamless integration with Claude Desktop
• Access to foundation models via Amazon Bedrock

Prerequisites

• Operating System: MacOS or Ubuntu (Nova Act requirements)
• Python 3.10 or higher
• A valid Nova Act API key (obtain from https://nova.amazon.com/act)
• Node.js (for Claude Desktop integration)
• Amazon Bedrock access:
  • Amazon Bedrock enabled in your AWS account
  • Claude 3.5 Sonnet V2 model enabled (this is default for this example)
  • AWS credentials and region properly configured - AWS CLI Quickstart Guide

Project Structure

The repository contains two main example directories:

1. Streamlit Examples (streamlit_examples/)

• video_game_research_st.py: A comprehensive tool that:
  • Finds top games for any selected gaming system
  • Searches Amazon in parallel for pricing and details
  • Creates interactive result tables
  • Saves research data for future reference

2. MCP Examples (mcp_examples/)

• nova_act_mcp_server.py: MCP-compliant server implementation exposing Nova Act capabilities
• nova_act_mcp_client.py: Example MCP client implementation
• Demonstrates integration with Claude Desktop and other MCP hosts

3. Nova Act Examples (nova_act_examples/)

• get_coffee_maker.py: Simple example demonstrating how to:
  • Initialize Nova Act with a starting web page