How MoEngage turned over 400 APIs into agent tools with AgentCore Gateway
MoEngage converted roughly 400 private REST APIs into permission‑scoped agent tools using Amazon Bedrock’s AgentCore Gateway, avoiding public endpoints and per‑team MCP servers. By centralizing the Model Context Protocol (MCP) layer in the AI Platform team, they reduced integration time from days to hours and saved an estimated six months of engineering effort. The approach enforces per‑agent authorization server‑side and maintains data residency across seven AWS Regions, enabling cross‑service agents without exposing internal services.
⚡ Key Takeaways
- ~400 internal REST APIs from 25 engineering teams across seven AWS Regions were exposed as agent tools via AgentCore Gateway.
- Centralized MCP layer owned by the AI Platform team replaces the need for >20 separate MCP servers.
- Per‑agent authorization is enforced server‑side, eliminating the need for public endpoints and reducing latency in tool discovery.
- Agents connect to a single tool surface; teams only need to provide an OpenAPI spec for their APIs.
- Onboarding a new API now takes hours instead of days, delivering a six‑month engineering effort saving.
Engineers can rapidly expose existing internal services to LLM agents without building custom MCP infrastructure, cutting integration time dramatically while preserving security and compliance.
✅ Practical Steps
- Register each internal API’s OpenAPI specification with AgentCore Gateway to generate permission‑scoped agent tools.
- Configure per‑agent authorization policies in the central MCP layer managed by the AI Platform team.
- Deploy agents to use the single AgentCore Gateway endpoint for tool discovery and execution, eliminating the need for public API exposure.
Want the full story? Read the original article.
Read on AWS ML Blog ↗