The MCP Gateway: A Journey from Idea to Infrastructure
summary
Mimecast builds an MCP Gateway to replace insecure local MCP servers, moving from a May–Sep POC to a December MCP server and a February production gateway that supports Claude, ChatGPT, OpenAI, Copilot and Gemini. It uses AWS Bedrock Agent Core Gateway, Amazon Cognito for OAuth with user-level consent, Spring AI for MCP servers, and an interceptor lambda to translate public JWTs into internal auth tokens. Result: end-to-end customer installs, audited calls, rate limiting and centralized authorization, and multi-product scaling.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:30Model Context Protocol
- 2:53Building MCP Gateway at Mimecast
- 5:19POC to production MCP server
- 8:51Local MCP server limitations
- 12:03MCP Gateway architecture
- 14:32Installing MCP Gateway
- 16:23Security and authorization
- 21:01MCP authorization design
- 23:09MCP Gateway lessons learned
- 25:36Introducing new ideas
- 27:22MCP future plans
- 29:50AI and software engineering
related talks
26mSecure Your MCP Servers with OAuth, JWT, SPIFFE and More
17mBuilding Production-Ready MCP Servers: From AI Agents to Secure Enterprise APIs.
21mMCP at 1. The state of the protocol you love (or hate)
27mSecuring the AI Revolution - [...]
12mBuilding Better Remote MCPs: Web3’s Answer to Auth & Monetization
26mGovernance and Security of APIs and MCPs
16mDriving agentic innovation w/ MCP as the backbone of tool-aware AI
29mBuilding Interoperable Agents: When A2A meets MCP