Intro to MCP

Shaw Talebi's video on MCP was a little better on explaining what the hype around MCP was:was versus the official docs by Anthropic:

Model Context Protocol (MCP) is a standardized protocol for AI software to talk to outside services and get output from them. It connects things like Copilot & Claude Desktop, Visual Studio Copilot, etc. to various tools and services for things like cloud services (Azure, AWS), version control, and so on.

MCP leverages client-server architecture. The servers expose different tools to clients. Any one client can have multiple server connections.

Clients discover what tools are on a server, receive server data, and manage tool execution.

Clients will be rare to develop (AI agents will have these, LLM-powered apps will have them), but MCP servers will be far more common to develop.

MCP servers expose prompts, resources (static data, a filesystem, one or more databases), and tools (custom methods, API access, image processing).

Servers can be spun up locally and can talk to clients over stdio, or they can be spun up remotely and talk to clients via HTTP requests.