Skip to content

Details

AI coding assistants like Claude Code, Cursor, and Copilot have changed how we write code -- but until recently, they couldn't securely touch or manage our cloud infrastructure. Giving an AI agent access to AWS APIs usually meant exposing highly privileged credentials to a local workspace, or risking multi-step hallucinations with no guardrails.

Enter the Model Context Protocol (MCP). In this technical, demo-driven session, we break down how to design, code, and productionize a custom
AWS-native MCP server -- moving from a brittle local prototype to a secure, enterprise-ready cloud architecture.

We'll build a Python MCP server with FastMCP, deploy it statelessly on Amazon ECS Fargate behind an Application Load Balancer, and lock it down with IAM task roles so the agent never touches a long-lived credential.

Then we'll connect it back to local IDEs like Claude Desktop and Cursor over OAuth 2.1 -- the same enterprise-auth path that lets identity providers like ADFS or IAM Identity Center sit behind it.

We close by looking at where the protocol itself is going next: the "Skills over MCP" effort, and how a community debate about what belongs in
the protocol turned into a shipped spec extension.

Related topics

You may also like