In this lesson, you'll learn how to make your agent dramatically more powerful by connecting it to external tool servers using the Multi-Capability Protocol (MCP).

MCP allows an agent to dynamically discover and use new tools without requiring any changes to its own code. We'll connect our agent to public servers that provide tools for searching AWS documentation, web search, and accommodation booking, turning our general-purpose agent into a specialized expert.

🚀 What You'll Learn

Dynamic Tool Discovery: How agents can discover and use new capabilities at runtime
Multi-Server Integration: Connecting to multiple MCP servers simultaneously
Real-time Data Access: Using external tools to access current information
Error Handling: Robust error handling for external service connections

📁 Files in This Lesson

main.py: Single MCP server integration with AWS documentation
multiple_mcp.py: Multi-MCP server integration with web search and accommodation tools

🔧 Key Concepts Explained

1. MCP Client (`MCPClient`)

The bridge between your agent and external tool servers. Configured to start servers using uvx or npx, which downloads and runs packages without manual installation.

2. Dynamic Tool Discovery

The magic happens inside the with mcp_client: block. The client:

Starts the server process
Connects to it via stdio
Discovers available tools using mcp_client.list_tools_sync()
Manages the server lifecycle

3. Agent with Dynamic Tools

Tools fetched from servers are passed directly to the Agent constructor. The agent learns about tools and their functions at runtime, without needing prior knowledge.

4. Tool-Powered Expertise

When asked questions, the agent's LLM brain:

Sees available tools from connected servers
Intelligently decides which tools to use
Finds accurate, up-to-date information
Avoids relying on potentially outdated training data

🛠️ Prerequisites

Required Environment Variables

# For the language model
NEBIUS_API_KEY="your_nebius_api_key"

# For web search capabilities (multiple_mcp.py)
EXA_API_KEY="your_exa_api_key"

Installation

# Install dependencies
uv sync

# Or with pip
pip install -r requirements.txt

🎯 Usage Examples

Single MCP Server (AWS Documentation)

# Run the AWS documentation agent
python main.py

Example Query: "What is the maximum invocation payload size for AWS Lambda?"

Multiple MCP Servers (Web Search + Accommodation)

# Run the multi-server agent
python multiple_mcp.py

Example Queries:

"What's the fastest way to get to Barcelona from London?"
"What listings are available in Cape Town for 2 people for 3 nights?"

🌟 Benefits of MCP

Extensibility: Add new capabilities without code changes
Maintainability: Update tools on servers, agents gain new abilities instantly
Modularity: Mix and match tools from different servers
Real-time Data: Access current information from external sources
Specialization: Transform general agents into domain experts

🚨 Error Handling

The enhanced code includes comprehensive error handling:

Configuration Errors: Missing API keys
Connection Errors: Network or server issues
Runtime Errors: Unexpected failures during execution

🔗 Further Learning

Watch the Video: AWS Strands Course Playlist
Read the Docs: Official Strands Documentation
MCP Protocol: Model Context Protocol

Previous Lesson	Next Lesson
Lesson 3: Structured Output	Lesson 5: Human-in-the-Loop

README.md

Lesson 4: Extending Agents with External Tools (MCP)