Agentic AI, APIs & MCP

The one-sentence version: a normal AI answers — an agentic AI gets things done. You give it a goal, and it figures out the steps, uses tools, checks its own work, and keeps going until the goal is met.

The analogy — a vending machine vs. an intern

A regular chatbot is a vending machine: put in a question, a snack pops out. One shot, done. An agentic AI is a new intern: say “Plan the office party,” and they make a to-do list, book the room, order pizza, text everyone, and come back when it's finished.

The loop that makes it "agentic"

The key idea: agents need tools to actually do things in the real world. That's exactly where APIs and MCP come in. 👇

Want the code behind this loop? The Week 1 deep dive takes it apart line by line and runs it live in your browser.

Theory: an API (Application Programming Interface) is a messenger that lets two pieces of software talk using agreed-upon rules. An API call is one round-trip: you send a request, you get back a response.

The analogy — a waiter at a restaurant

You (your app) are the customer. The kitchen is a giant server you can't walk into. The API is the waiter: you give your order (the request); the waiter brings your food (the response). You never need to know how the kitchen cooks — just how to order.

A real example — weather

Your weather app doesn't own a satellite. It makes an API call:

REQUEST:   GET weather for Austin, TX

RESPONSE:  { "temp": 98, "sky": "sunny", "humidity": 40 }

The reply comes back as tidy data (usually JSON, like above). Every refresh is another API call. You already use them constantly: “Log in with Google,” a live game leaderboard, or Google Maps inside Uber — all API calls.

On the Build It page you'll create this exact weather API yourself in 6 lines.

Theory: MCP is a universal standard that lets an AI connect to tools, apps, and data — all in the same way. Instead of a custom connection per tool, MCP gives every tool one shared plug shape.

The analogy — USB-C

Remember when every phone had a different charger? Then USB-C arrived — one plug for laptop, phone, headphones, monitor. Before MCP, every AI-to-tool link was a custom cable (Gmail? build one. Calendar? build another). With MCP, tools expose themselves through one standard port, so any MCP-speaking AI plugs in instantly. Build it once, works everywhere.

They're not rivals — MCP usually uses APIs underneath. But they solve different problems.

No coding required to start. Three tiers, easiest first.

"Search the web for today's top science headline,
 then make me an image about it."