# MoEngage uses AgentCore Gateway to expose 400 private APIs as agent tools

published: 2026-10-30T14:05:45.667Z
topic: mcp
source: https://aws.amazon.com/blogs/machine-learning/how-moengage-turned-over-400-apis-into-agent-tools-with-agentcore-gateway/
url: https://aiagentnews.io/stories/moengage-uses-agentcore-gateway-to-expose-400-private-apis-as-agent-tools

MoEngage, the customer engagement platform, used Amazon Bedrock AgentCore Gateway to turn about 400 private internal REST APIs into permission-scoped MCP tools for its agents. The company says the approach saved an estimated six months of engineering effort and cut the time to onboard a new API from days to hours.

MoEngage runs roughly 400 internal REST APIs built by 25 engineering teams, hosted in private VPCs across seven AWS Regions to meet data residency commitments. Rather than have each team run its own MCP server, which it judged would mean more than 20 deployments to maintain, MoEngage assigned the MCP layer to a central AI Platform team. Product teams now maintain OpenAPI specs and mark which operations should become tools.

AgentCore Gateway generates MCP tools from registered OpenAPI specs, with each tool named after its target and operation ID. Agents sign requests with SigV4 using their IAM roles, and Cedar policies decide which tools each agent can list and call before the catalog is returned. Gateway traffic reaches internal services through managed egress over Amazon VPC Lattice, and inbound access goes only through a VPC interface endpoint.

Tenant, user and role context is passed in headers set by the client library, so the model cannot set or change it. A search_tools meta-tool lets an agent find relevant tools by task description, so the full catalog never enters the context window. MoEngage has run ten internal agents on this setup in production for three months.

## Why it matters if you build agents

Teams building agents on internal APIs can avoid running a separate MCP server per service by generating tools from existing OpenAPI specs behind one gateway. Server-side Cedar policies keep write operations out of an agent's tool list entirely, which is stronger than filtering tools on the client. Keep identity out of tool parameters so the model cannot set or override tenant or user context.