Why this term shows up everywhere in this channel Model Context Protocol (MCP) appears here in two settings: an episode with Stainless founder Alex Rattray exploring how to build and scale it, and a conversation with Microsoft CTO Kevin Scott about the infrastructure agents need. Both describe the same underlying problem — the internet was architected for a pre-AI world, and agents need a way to reach it. How MCP is used in these conversations - In the Stainless episode, MCP is introduced as the technology that turns any website or web service into a set of tools an AI can use natively to get work done. The host argues that software companies who do MCP well are positioned to win over the next decade, and the conversation digs into the nitty-gritty: how to design good MCPs, why they are hard to scale and possibly insecure, and what a better model for letting AIs use the internet might look like. Stainless itself builds APIs and SDKs for major AI companies and is now building MCP servers. MCP as infrastructure - In the Kevin Scott conversation, the same idea appears as the push for a standard protocol: a plumbing layer, like core internet protocols, that lets agents talk to the websites and APIs that already exist — with Scott pushing for all of Microsoft's internal systems to speak such a protocol to every agent the company writes. - MCP's job in this material: let AI agents use websites and web services as native tools. - It is framed as a bridge between a pre-AI internet and agentic software. - Open problems include scaling and security — a better model is still unsolved. - The same idea surfaces as a standard protocol push at Microsoft for agent-facing plumbing.