MoEngage used Amazon Bedrock AgentCore Gateway to convert roughly 400 private REST APIs spread across 7 AWS Regions into permission-scoped tools for AI agents. According to the post, the approach avoided building per-service MCP servers, exposing public endpoints, or overloading agent context with 400 separate tools. The implementation combined OpenAPI specifications, Cedar policies for permissions, and managed Amazon VPC Lattice for egress.
Why it matters
Exposing large numbers of internal APIs as agent tools raises challenges around access control, network exposure, and context size. The described setup addresses these by scoping permissions and keeping endpoints private while still making the APIs usable by agents.
Who should care
Teams building AI agents on top of existing internal API estates, particularly those operating across multiple AWS Regions and already using Amazon Bedrock, may find the pattern relevant.