How It Actually Works · Episode 1
How MCP actually works
One plug lets any AI app use any tool, and you decide what runs.
The Model Context Protocol in 85 seconds: why AI apps used to need glue code for every tool, what the one plug is, why there is no handshake, the three things a server offers, and who stays in charge.
One tool call, step by step
Transcript
0:00What are we learning today?
What are we learning about today?
MCP. The Model Context Protocol.
0:15Before MCP: glue code everywhere
Before MCP, every AI app needed its own custom connector for every tool. GitHub, your database, Slack: each one wired by hand.
0:25One plug, and no handshake
MCP is one standard plug. Any AI app that speaks it can use any tool that speaks it. The app is the client. The tool is the server.
And there’s no handshake. Every request carries its own label: which version it speaks and what it can do. So the server can answer each one fresh.
0:43Prompts, resources and tools
Behind the plug, a server offers three things. Prompts: ready-made templates you choose. Resources: files and data the app hands over as context. And tools: actions the AI decides to call, like opening a GitHub issue.
0:57You decide what runs
Tools change real things. So the app checks with you first: you approve each call, or approve it ahead of time. Either way, you decide.
1:04A real one: GitHub’s MCP server
This is a real one. GitHub’s MCP server, and its tools are right here.
One plug. Any tool. But how does the server know it’s really you? Next time: OAuth.
Sources
Next: Episode 2, How OAuth actually works →
The narration is an AI voice, and the animation was made with AI tools.
The robot animation is made with Kling. Try Kling with my referral link. It’s a referral link, so signing up through it may earn me credits.
The narrator’s voice is made with ElevenLabs. Try ElevenLabs with my referral link. It’s an affiliate link, so signing up through it may earn me a commission.