2-hour hands-on workshop

Build Your First Agentic AI

Agentic AI · APIs · MCP — and by the end, you design your own.

Today's map

Where we're headed

🤖 Agentic AI 🍽️ API calls 🔌 MCP ⚔️ API vs MCP 🎒 A real use case 🧭 Full build walkthrough 🛠️ Build your own

One idea ties it all together ⤵

A goal → a Think·Act·Check loop → real tools

Everything today is free · no Claude account

Our 3 tools

🧠

Google AI Studio

The brain — chat with Gemini, free.

🔌

Groq

The API you call to reach an AI.

🧩

Flowise

Where you build your own agent.

Brain → API → build.

The big idea

🥤 Vending machine

Press a button → a snack drops. One press, one result. Done.

vs

🧑‍💼 An intern

"Plan the party." They pick a date, order pizza, text everyone, come back when it's handled.

Today we build the intern.

Part 1

Agentic AI

You give a goal. It figures out the steps, uses tools, checks its work, and keeps going.

The engine of every agent

Think→ Act→ Check→ Repeat

…until the goal is done.

Remember this

A brain can only think.
Tools are its hands.

No tools = no agent. That's the whole reason APIs and MCP exist.

Part 2

API Calls

The waiter between your app and the kitchen: you send a request, you get a response.

Both sides of one call Both sides of an API call, request out and response back

Your inputs go inside the request → server looks up the answer → sends it back as data.

✋ Hands-on #1 · round 1 · warm-up (no key)

Paste this in your browser. Hit Enter.

https://api.open-meteo.com/v1/forecast?latitude=30.27&longitude=-97.74&current=temperature_2m

That blob is the response — exactly what a weather app receives. That was a GET: "give me data."

✋ Hands-on #1 · round 2 · a real call to an AI (Groq)

In hoppscotch.io → method POST → this URL:

https://api.groq.com/openai/v1/chat/completions

# Headers
Authorization: Bearer <class key on the board>
Content-Type: application/json

# Body (raw JSON)
{ "model": "llama-3.1-8b-instant",
  "messages": [{ "role":"user",
    "content":"Tell me a joke about high school." }] }

Request carries your message + a key (your ID badge). Response = JSON. You just called an AI through an API.

Part 3

MCP = USB-C for AI

One universal plug so an AI connects to many tools the same way. Build once, works everywhere.

Part 4

API vs MCP

An API is a door.
MCP is a universal door frame — so an AI opens any door the same way.

…and behind many doors, an API still does the work.

☕ Break's over · lightning quiz

Chatbot vs agent — one difference? (uses tools / multi-step)
Who's the API in the restaurant? (the waiter)
Request vs response? (the order / the food)
MCP is like what plug? (USB-C)
Does MCP replace APIs? (no — it uses them)
The 4-step loop? (Think·Act·Check·Repeat)

Part 5 · a use case you'll relate to

The Study-Buddy Agent

A stressed student and a friendly study agent

You bring the goal. It brings the plan.

🗣️ "I have a bio test Thursday. Help me be ready."

Think — 4 days, ~5 topics, practice Tue & Thu

Act — reads notes, builds a 3-day plan

Act — drops study blocks in your calendar

Act — quizzes you, spots you miss osmosis

Check — nailed it? move on. missed it? re-drill

Repeat — adjusts daily until test day

🚀 5 agentic AI ideas you could build

📚

Study-Sprint Coach

Quizzes you, tracks weak topics, drills them before a test.

🎬

Club / Event Planner

Finds a date, makes a flyer, messages members, books the room.

🎓

College-App Wrangler

Tracks deadlines, checks requirements, nags you about what's left.

💸

Money & Shifts Tracker

Logs shifts, sorts spending, says if you can afford it.

🗣️

Language Partner

Chats at your level, fixes mistakes, remembers your weak words.

🏀

Practice Buddy

Builds a drill plan, adjusts to last session, logs progress.

Part 6 · full walkthrough

The Group-Project Rescuer

Every agent goes through these same stages. None need a CS degree.

Stage 0

The Itch

A real, annoying problem you personally have. "Nobody tracks who's doing what."

Stage 1

The Goal

ONE sentence: "Split tasks, remind each person daily, warn me if we're behind."

Stage 2

The Tools

Task list · calendar · a messenger (Discord/email). Each reached by an API / MCP.

Stage 3

The Loop

Think: who's behind? → Act: nudge them → Check: anyone done? → Repeat: every morning.

Stage 4

Pick your pieces

🧠 The brain — an AI model that decides what to say

✋ The hands — MCP connectors to Discord + a Google Sheet

🖥️ A place to run — your laptop now; a free scheduler later

Stage 5 · Build the MVP

The smallest version that works (~10 lines)

tasks = read_sheet("project_tasks")      # the TOOL
days_left = deadline("Friday") - today()

for person in tasks:
    if person.status != "done":
        msg = ai("Nudge " + person.name +
                 ", part due in " + days_left + " days")   # THINK
        send_discord(person.name, msg)   # ACT

if unfinished(tasks) > 2 and days_left <= 2:
    send_discord("YOU", "⚠️ 3 tasks open, 2 days left!")   # CHECK

Stage 6

Watch It Run

Test by hand → schedule it for 8am daily → now it works while you sleep.

Stage 7

Level Up

One improvement at a time. Each upgrade = one more tool, one more line.

The universal recipe

Itch → one-sentence goal → tool list → Think·Act·Check loop → tiniest version that runs → level up.

✋ Hands-on #2 · Phase A · design

Design it on the Canvas

1. Pick a case (yours or from the 5)

2. The itch → the goal (ONE sentence)

3. List 2–3 tools · label each API or MCP

4. Write the Think·Act·Check loop

5. Write your system prompt ("You are a ___ that ___")

✋ Hands-on #2 · Phase B · build it for real

Make it real in Flowise 🧩

1. cloud.flowiseai.com → Add New → Chatflow

2. Drag GroqChat · paste class key · model llama-3.1-8b-instant

3. Drag a Conversation Chain

4. Wire GroqChat → the chain's Chat Model

5. Paste your system prompt · Save · 💬 chat!

Groq = the brain · your prompt = the goal · each message = the loop.

Share-out · 45 sec each

Pitch it in one breath

"Our goal is ___.
Its tools are ___.
Its loop is: think ___, act ___, check ___."

📌 Recap in one breath

Agentic AI — goal + Think·Act·Check loop + tools
API call — a waiter carrying request & response
MCP — USB-C for AI: one plug, many tools
API vs MCP — a door vs a universal door frame

This week's challenge

Give your agent a hand. 🖐️

Add ONE tool to your Flowise agent so it can do something, not just talk. Text me what it did.

🧠 Google AI Studio🔌 Groq🧩 Flowise

You're a builder now.

Goal → loop → tools. Go make a hundred more.