How MoEngage turned over 400 APIs into agent tools with AgentCore Gateway

MoEngage used Amazon Bedrock AgentCore Gateway to turn roughly 400 private REST APIs into permission-scoped agent tools, saving an estimated six months of platform engineering effort.

• The customer engagement platform's APIs are built by 25 engineering teams and run in private VPCs across seven AWS Regions, isolated regionally to meet data residency commitments and never exposed to the public internet.
• MoEngage rejected giving each team its own MCP server, since more than 20 servers would mean more than 20 deployments to monitor and patch, and cross-domain agents would have to connect to several servers.
• Instead, teams build REST APIs and declare which operations should be tools, while the AI Platform team owns the MCP layer end to end.
• MoEngage set four constraints: no public endpoint, identity injected by the calling service rather than set by the model, and a tool catalog that does not flood the context window.
• At roughly 50 tokens of JSON schema per tool, a full tools/list across the estate is over 20,000 tokens on every connection, and client-side filtering does not fix it because the full catalog has already crossed the wire.
• Onboarding a new API now takes hours instead of days, with no MCP server code written by product teams and no public endpoint in the path.