Understanding IAM for Managed AWS MCP Servers

This article explains how to use IAM to govern AI agent access through AWS-managed Model Context Protocol (MCP) servers with standardized security controls.

• Two new IAM context keys enable differentiation between AI and human-driven actions
• aws:ViaAWSMCPService (boolean) allows denying all MCP-initiated actions organization-wide
• aws:CalledViaAWSMCP (string) restricts actions to specific MCP servers like EKS or ECS
• Simplified authorization model eliminates need for separate MCP-specific IAM actions
• VPC endpoint support coming soon for private network communication in regulated industries
• Defense-in-depth approach combines network perimeter and service-level IAM controls
• Start with restrictive policies and monitor CloudTrail logs to refine access over time

AWS-managed MCP servers now integrate with existing IAM policies, allowing organizations to secure AI workflows while maintaining flexibility for governance and compliance requirements.