The AWS Machine Learning Blog outlines an architecture for building an AI agent that spans multiple AWS accounts. In this design, a central platform account hosts the agent using Amazon Bedrock AgentCore Gateway and the Model Context Protocol (MCP), while individual line-of-business accounts expose their own data as MCP servers. This lets the agent query across accounts through a unified interface while each team retains its data within its own account.
The approach relies on secure cross-account access and fine-grained authorization to control how the agent reaches data in other accounts.
Why it matters
Many organizations separate data by team or business unit across distinct AWS accounts. This pattern shows how AI agents can query across those boundaries without consolidating the data, preserving account-level isolation while enabling unified access.
Who should care
Teams building AI agents on AWS, cloud architects, and platform engineers working with multi-account setups and Amazon Bedrock AgentCore may find the pattern relevant.