Run custom MCP proxies serverless on Amazon Bedrock AgentCore Runtime

This article explains how to deploy a serverless MCP proxy on Amazon Bedrock AgentCore Runtime to add custom controls to AI agent-tool interactions.

• MCP proxy acts as intermediary between agents and upstream MCP servers with custom logic
• Proxy dynamically discovers tools from upstream server and re-exposes them transparently
• Supports both IAM and OAuth JWT authentication methods for upstream server connections
• Authorization enforced independently at each layer: agent-to-proxy, proxy-to-upstream, upstream-to-tools
• Enables customizations like PII tokenization and tool-level access control without modifying upstream server
• Automated deployment script handles IAM role creation, container building, and AgentCore Runtime deployment
• Proxy runs stateless on AgentCore Runtime with automatic scaling and CloudWatch observability

The solution provides a reusable pattern for organizations with existing MCP filtering logic to add governance controls at the protocol layer while maintaining separation from upstream tool execution.