HEMA builds internal AI assistant HAL on Amazon Bedrock AgentCore, bringing knowledge into everyday tools through MCP

AWS published a HEMA implementation case study on September 23, 2026, describing how the Dutch retailer—more than 100 years old and operating over 750 stores across multiple countries—built its internal AI assistant, HAL, on Amazon Bedrock AgentCore. HEMA says employees previously had to search disconnected wikis, service catalogs, and IT portals. Some answers that once required checking three or four portals and could take an afternoon are now available in seconds from a chat window or integrated development environment.

HAL consolidates HEMA technology knowledge, including its service catalog, API specifications, documentation, and how-to material. Model Context Protocol (MCP) makes that knowledge available through the HAL chat interface and tools such as Kiro and Claude. The approach lets different clients use the same knowledge and tools instead of requiring a separate integration for each one. AWS says HAL now serves product owners and business analysts as well as developers.

For access control, the external MCP gateway authenticates through Microsoft Entra ID, so external clients do not need AWS credentials. Knowledge access is currently read-only and uses existing Active Directory group permissions. HEMA says it tested HAL in staging with engineers and business users for one month before wider production adoption.

HEMA describes the next step as extending HAL from a read-only knowledge assistant into an assistant that can carry out actions, citing an AWS account request made through chat or another client as an example. The post presents this as a future direction, not a feature already launched. In related AWS context, a September 24 guide described using AgentCore Gateway and MCP for an agent to query data across AWS accounts while keeping the data in the account that owns it.