The Game Master (GM) plays a central role in role-playing games: they create the players’ stories, embody non-player characters (NPCs), and ensure the rules are followed. AI can be a powerful ally for GMs, helping them check rules, improvise scenarios, and create content on the fly. In this talk/live coding session, we’ll build together an AI multi-agent Game Master using Spring AI. We’ll set up specialized agents for different GM responsibilities and explore how they collaborate through the Agent-to-Agent (A2A) protocol and tool integrations via the MCP protocol.

Devoxx

A live conference talk walkthrough building a multi-agent AI game master for tabletop RPGs using Spring AI, Java 25, and Amazon Bedrock. The demo progresses through five chapters: setting up a basic AI agent with a system prompt, adding built-in web-fetch tools, converting existing Java functions into agent tools using @Tool annotations, exposing tools over the network via an MCP (Model Context Protocol) server, and finally connecting specialized agents (character agent, rules agent) using the Agent-to-Agent (A2A) protocol. Key concepts covered include MCP for tool discoverability, A2A for agent-to-agent communication, Spring Boot for hosting MCP servers, and using Amazon Bedrock's Converse API to swap between LLM providers. The final architecture features an orchestrator game master delegating to specialized agents, with dice rolling exposed as an MCP tool.

Coding a Multi-Agent Game Master with Spring AI by Alessandra Pasini and Arnaud Jean