MoEngage has connected roughly 400 internal REST APIs to AI agents through Amazon Bedrock AgentCore Gateway while keeping the services inside private networks. The system turns approved API operations into permission-scoped tools without requiring each product team to build and maintain its own Model Context Protocol server.
The APIs span 25 engineering teams and seven AWS regions. They handle functions such as campaign creation, segment queries and delivery statistics, with regional isolation used for data-residency commitments. MoEngage rejected an architecture with more than 20 separate MCP servers because cross-domain agents would need several connections and every team would inherit deployment, patching and protocol work.
Instead, product teams continue to publish OpenAPI specifications and declare which operations should become tools. A central AI platform team owns the gateway. Server-side authorization limits which agent can call each operation, while managed private connectivity avoids opening the underlying APIs to the public internet.
The company estimates that this design avoided about six months of platform engineering and reduced onboarding for a new API from days to hours. Those estimates come from MoEngage and AWS, but the architecture illustrates a broader shift: agent deployment is increasingly about identity, policy and tool governance rather than simply choosing a model.