From portal-hopping to instant answers: HEMA’s journey with MCP and Amazon Bedrock
aws ·
HEMA, a century-old Dutch retailer, replaced fragmented internal knowledge retrieval with HAL, an AI assistant built on Amazon Bedrock AgentCore and the Model Context Protocol (MCP). Previously, engineers struggled with 'portal-hopping' across disconnected wikis and catalogs to find procedural or infrastructure information. HAL consolidates this data into a governed source of truth, utilizing semantic search and live API integrations via OpenAPI specifications. The solution leverages MCP to deliver instant answers directly within daily tools like IDEs and chat applications, eliminating the need for custom per-tool integrations. Security is anchored in Microsoft Entra ID, allowing external clients to access the system without exposing AWS credentials through a specialized authentication proxy. Initially deployed as a standalone web chat, HAL expanded to support multiple MCP-compatible clients, serving developers, product owners, and business analysts alike. This read-only knowledge layer has proven effective, setting the foundation for a future action layer where HAL will execute operational tasks such as provisioning resources.
aws