AWS Builder Center

Builder Center

Building My First Autonomous AI Agents with Amazon Nova Pro and MCP

My journey through the AWS User Group Madurai – Nova Month Challenge. I built five progressive AI agents from scratch using Amazon Bedrock, the Strands Agents SDK, and the Model Context Protocol (MCP), culminating in a fully automated web research assistant.

The Builders Skill Sprint: My AI Agent Journey

As generative AI evolves from simple chatbots to autonomous systems, the need to understand how to build AI Agents has never been greater. Recently, I participated in the AWS User Group Madurai – Nova Month Challenge, an intensive skill sprint designed to teach developers how to build modern AI agents step-by-step.
If you want to jump straight to the code, check out my GitHub Repository here 
In this post, I want to share my journey of building these agents using Amazon Nova Pro, the Strands Agents SDK, and the powerful Model Context Protocol (MCP).

The Progression: 5 Challenges to Autonomy

The skill sprint was brilliantly structured. Instead of jumping straight into complex systems, we built our understanding progressively across 5 challenges:
  1. The First AI Agent: Starting simple with Ollama and the Strands SDK to understand the basic agent loop.
  2. AI Agent with Tools: Moving to Amazon Bedrock (Nova Pro) and giving the agent custom tools using the @tool decorator, allowing it to fetch live weather and calculate dates.
  3. Persistent Memory: Teaching the agent to remember facts across sessions using mem0 and local FAISS vector databases.
  4. The Full Agent: Combining streaming, persistent memory, and dynamic tool calling into a single cohesive agent.
  5. The Capstone: Integrating MCP (Model Context Protocol) to give the agent secure, zero-code access to external systems.

The Highlight: Building "The Automated Researcher"

For my final capstone project (Challenge 5), I wanted to build something genuinely useful: The Automated Researcher.
Unlike traditional LLM chats where you have to manually copy-paste search results, this agent is an autonomous research assistant. You give it a topic, and it:
  1. Searches the live web for up-to-date information.
  2. Synthesizes the scattered findings into a cohesive, structured study guide.
  3. Automatically saves the markdown guide directly to a local directory on my machine.

How it Works: The Magic of MCP

The secret sauce behind this agent is the Model Context Protocol (MCP). Instead of writing custom API integrations for web searching and file management, I used pre-built MCP servers:
  • DuckDuckGo MCP Server: Provided the agent with instant, zero-configuration web search capabilities without needing API keys.
  • Filesystem MCP Server: Allowed the agent to create directories and write markdown files, strictly sandboxed to a specific research_notes/ folder for security.
By simply pointing the Strands SDK to these MCP servers, the Nova Pro model instantly understood how to use them.
1
2
3
4
5
6
7
python# Connecting the agent to the web and filesystem via MCPddg_mcp = MCPClient(
lambda: stdio_client(StdioServerParameters(
command="npx", args=["-y", "@oevortex/ddg_search@1.2.2"]))
fs_mcp = MCPClient(lambda: stdio_client(StdioServerParameters(
command="npx", args=["-y", "@modelcontextprotocol/server-filesystem",
"./research_notes"])))agent
= Agent(model="us.amazon.nova-pro-v1:0", tools=ddg_mcp.tools + fs_mcp.tools)

Key Takeaways

Building with Amazon Bedrock and the Strands SDK was an eye-opening experience. Here is what I learned:
  1. Nova Pro Excels at Tool Use: Amazon Nova Pro is incredibly reliable at interpreting tool schemas and executing multi-step reasoning workflows.
  2. MCP is the Future of Integrations: The Model Context Protocol is a game-changer. It standardizes how AI models connect to data sources and tools, eliminating hours of boilerplate code.
  3. Sandboxing is Critical: When giving an AI agent the ability to write to your filesystem, strict sandboxing (like restricting the Filesystem MCP to a specific folder) is absolutely essential.

What's Next?

This skill sprint was just the beginning. Moving forward, I plan to explore integrating MCP with advanced RAG systems to build enterprise-grade document analysts, and eventually deploying these agents as scalable serverless applications on AWS.
If you want to see the complete code for all 5 challenges, check out my GitHub Repository .
Keep building! 🚀
# generative-ai# amazon-bedrock# python# ai-agents
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article