As AI moves beyond static RAG to dynamic, goal-oriented agents, developers face a critical challenge: how to securely connect natural language models to complex enterprise data without managing heavy infrastructure.

In this session, I'll dive deep into a production-ready architecture that combines Elastic’s Model Context Protocol (MCP) with the Amazon Bedrock AgentCore Runtime. I'll demonstrate how to transform Elastic from a search tool into a "context engine" that translates plain English like "find high-risk investment opportunities" into precise technical queries, all orchestrated within a serverless AWS environment.

Will be covering the end-to-end technical implementation:

• Amazon Bedrock AgentCore: AgentCore solves the "stateless" problem by injecting Mcp-Session-Id headers to maintain conversation context for isolation and providing robust memory management for your agents.

• Elastic MCP + Agent Builder: How to containerize the Elastic MCP server (using Docker and Amazon ECR) to enable dynamic "tool discovery," allowing agents to query logs, documents, and vector stores intelligently.

• Security & Scale (high-level): Moving away from API keys to enterprise-grade security using AWS IAM service roles and SigV4 authentication, ensuring your agents are secured.