AWS re:Invent 2025 - Build a multi-agent role-playing Game Master with Strands Agents (DEV330)

Building a Multi-Agent Role-Playing Game Master with Strands Agents

Overview

The presentation showcased how to build a multi-agent role-playing game master using the Strands Agents framework. The goal was to create an AI-powered Dungeon Master assistant to help manage the complexities of running a tabletop role-playing game like Dungeons & Dragons.

Key Components

The system consisted of several specialized agents and tools working together:

1. Orchestrator Agent

• The main "Game Master" agent that coordinates the other components
• Relies on other agents and tools to perform various tasks

2. Character Management Agent

• Responsible for creating and managing player characters
• Stores character data in a file-based "database"

3. Rules Agent

• Provides access to the game's rule book to ensure actions are valid
• Uses a vector database to quickly look up and apply game rules

4. Dice Rolling Tool

• Provides a standardized way to roll dice for the game
• Can be exposed as an MCP (Model Context Protocol) server for easy access

Technical Implementation

• The presentation used the Strands Agents SDK to quickly build the multi-agent system
• Strands Agents supports both MCP (for connecting to tools) and A2A (Agent-to-Agent) protocols
• Agents can be configured to use different language models (e.g. Amazon Bedrock, Anthropic, etc.)
• Tools are defined using Python functions and decorated to expose them to the agents
• Agents communicate through prompts, with the orchestrator delegating tasks to specialized agents

Business Impact

• Automating the role of the Dungeon Master can make tabletop RPGs more accessible
• AI-powered game masters can provide a consistent, knowledgeable experience for players
• The modular, multi-agent architecture allows for easy extensibility and customization
• Potential applications beyond gaming, such as virtual assistants, customer service, and more

Demonstration

The presenters walked through a live coding demo to build the core components of the game master system:

1. Created a basic agent that can use different language models
2. Added tools to the agent, such as a "current time" function
3. Exposed a custom "roll dice" function as an MCP server
4. Built a separate "character management" agent to handle character creation and storage
5. Integrated the orchestrator agent to coordinate the various components
6. Demonstrated the end-to-end workflow of creating a new character and handling game actions

Key Takeaways

• Strands Agents simplifies the development of multi-agent systems with its SDK and protocol support
• Modular, specialized agents can be combined to create powerful AI assistants
• Exposing functionality as MCP servers enables easy integration and reuse
• Prompt engineering and system prompts are crucial for guiding agent behavior
• The presented system showcases the potential of AI-powered game masters and virtual assistants