Build a multi-account AI agent with AgentCore Gateway and MCP

For organizations grappling with the secure, scalable deployment of AI agents across diverse data silos, a new approach promises unified access without compromising data sovereignty. The AWS Machine Learning team recently detailed a method for constructing multi-account AI agent architectures using AgentCore Gateway and the Multi-Account Control Plane (MCP). This innovation addresses the complex challenge of allowing AI agents to query data residing in separate AWS accounts—each potentially owned by different teams or departments—while maintaining stringent security and access controls. Essentially, it centralizes the AI agent's operation in a platform account, enabling it to interact securely with data exposed by MCP servers within individual line-of-business accounts, ensuring fine-grained authorization and cross-account access. This technical pattern significantly impacts how larger enterprises, or even those experiencing rapid growth, can leverage AI agents without consolidating all their sensitive data into a single location. Consider a mid-sized financial services firm in New York City: their compliance, trading, and customer service departments each hold proprietary data in isolated environments. An AI agent built with this architecture could analyze customer interactions for compliance risks, cross-reference trading data for anomalies, and personalize service responses—all without any single team's data ever leaving its designated secure account. For a logistics startup headquartered in Chicago, managing vehicle telemetry, supply chain manifests, and customer delivery preferences, this means an AI agent could optimize routes and predict delays by securely accessing disparate datasets, improving efficiency and customer satisfaction without data migration overhead. An internal IT team at a manufacturing company in Detroit could deploy an AI-driven helpdesk that queries operational data from factory floor systems and employee HR records—both kept strictly separate—to resolve issues faster and more intelligently. The practical advantage lies in minimizing data movement, reducing security risks, and simplifying compliance while maximizing the utility of AI agents. You can capitalize on this by designing your next AI agent deployment with a secure, distributed data strategy from the outset, rather than attempting to retrofit it later. This architectural pattern offers a path to build more robust, scalable, and secure AI applications, particularly beneficial for those subject to strict data governance or regulatory requirements. To begin exploring this, consider an internal use case at your own organization where an AI agent could provide value by accessing information spread across two distinct data sources, each currently residing in its own isolated environment. Map out how these data sources could expose relevant information via an MCP server in a separate AWS account, then sketch a high-level architecture for a central agent account to orchestrate queries. This small exercise can illuminate the potential for secure, multi-source AI integration within your existing cloud footprint.