From Talking to Doing: Why Model Context Protocol MCP Are the Next Big Step for AI
- Hind Khayati
- Aug 15
- 6 min read
August 7, 2025 Guest Author: Hind Khayati When AI Learns to Act: An Introduction to Model Context Protocol (MCP)
Before, our go-to move was Googling the answer whenever we needed information. Now? We turn to LLMs. Whether it’s deciding what to cook for dinner, fixing a tricky line of code, or figuring out how to balance training for a marathon while learning a new skill, it’s become second nature to just ask an AI.
But here’s the thing: until recently, most AI models lived in their own bubble. They could talk, but they couldn’t act. They didn’t know what was in your files, they couldn’t check the latest numbers, and they definitely couldn’t go into your tools to get something done.
During my apprenticeship as a data scientist, I kept hearing a new term pop up in conversations: MCPs (Model Context Protocols). At first, it was just another acronym I’d skim past in articles and GitHub repos. But my curiosity got the better of me. I dug deeper, and the more I learned, the more I thought: I need to try this myself.
What Are MCPs, Really?
Think of MCPs as a universal translator and an access pass for AI. They give models a standard, secure way to connect to external tools, databases, and services so they can pull in live, relevant context before answering or taking action. Technically, this often involves the AI using a predefined set of APIs and data schemas, ensuring it speaks the same language as the tools it connects to.
Without MCPs: your AI is like a brilliant person locked in a room without internet access.
With MCPs: that same person can step outside, talk to the right people, check the latest data, and come back with a clear, informed answer—or even complete the task for you.
Source: A.I News Hub
Why MCPs Matter
As someone who loves experimenting, I see MCPs as a gateway to a new kind of AI workflow:
Instead of manually looking up and organizing information, I can let my AI pull exactly what it needs from the source.
Instead of switching between apps to get something done, I can stay in one conversation and let the AI do the switching.
Imagine telling your AI:
“Find all my flight options for next weekend, compare prices, and book the cheapest one that lands before 10 a.m.”
With MCPs, that’s not a multi-step juggling act—it’s a single, natural-language request.
For me, this is where it gets really exciting. It’s about more than just convenience; it’s about reducing the 'cognitive tax' we pay every day. Think about how much mental energy is spent just remembering which app holds which piece of information. MCP promise a future where I don’t have to be the project manager for my own life. My AI can handle the logistics, freeing up my mind to focus on the things that actually matter, like being creative, solving bigger problems, or just being present
A Vivid Example: How It Works
Let's use that marathon training example from the intro. It’s no longer a theoretical question for the AI; it becomes a practical planning session.
You ask.
“How should I adjust my marathon training this week, given my schedule and last week's performance?”
The AI connects through MCP.
It accesses your calendar for free time, connects to your fitness app (like Strava or Garmin) for your recent run data (pace, distance, heart rate), and queries a weather API for the forecast.
It retrieves live context.
It sees you have a long meeting on Wednesday, your pace was slower on your last long run, and it's going to rain on Friday.
It acts with intelligence.
Instead of a generic plan, it replies: "Let's move your long run to Saturday to avoid the rain. We'll do a shorter recovery run on Wednesday after your meeting. Focus on a slower pace this weekend to recover properly."
My First Simple MCP Experiment
True to my word, I couldn't just read about it. I had to build something, even if it was small.
My goal was simple: connect my to-do list app (Todoist) to an AI so I could ask, "What are my top 3 priorities for today?" and have it pull the tasks directly from my personal board.
Using a simple Python script and the app's API key, I defined a "function" the AI could call. This function told the AI exactly how to authenticate and which data to request. It was a bit clunky, but the first time I asked the question and saw the AI fetch my actual, live to-do list in response was a huge 'wow' moment. It made the abstract concept of MCPs completely real and showed me the power in even the simplest connection. That little success completely reframed my perspective as an apprentice. Suddenly, the complex data pipelines and workflows I was studying seemed less daunting. I started to see how MCPs could act as the connective tissue, allowing a data model to not just analyze historical data, but to actively query live production databases, ping a teammate on Slack for clarification, and update a project dashboard, all from a single prompt. It’s the missing link between data science and data action.
Real-World Use Cases
Once you see it work, the possibilities feel endless:
Travel: Find your booked hotel, check the check-in time, and add it to your calendar automatically.
Home Life: Monitor your grocery list, order missing items online, and schedule delivery.
Health & Fitness: Pull your step count, check your gym’s class schedule, and reserve a spot—all in one go.
Work: Summarize unread emails from your manager, draft a reply, and schedule a meeting if needed.
Challenges and What’s Next
MCPs are still new, and like any early technology, they have hurdles:
Security: We’re giving AI access to real accounts. This means getting granular with "scoped permissions"—giving the AI just enough access to do its job, and nothing more. Audit trails are critical.
Adoption: Not every app supports MCP yet. The growth of open-source communities building shared connectors will be key to overcoming this.
Latency: Connecting to multiple sources can be slower, so optimization is essential.
But looking ahead, I see a huge opportunity:
Personalized MCP setups where my AI knows my habits—from my morning coffee order to my vacation preferences.
AI assistants that collaborate through MCPs to coordinate big tasks, like planning a wedding or moving to a new city.
Open-source MCP libraries that make experimenting easy for anyone curious.
Final Thoughts
For me, learning about MCPs feels like seeing AI take its next big step—from being a smart conversationalist to becoming an actual teammate.
It’s exciting to imagine the prototypes anyone could build:
An AI that automatically checks the weather and packs your suitcase list for you.
An AI that keeps track of all your subscriptions and cancels the ones you don’t use.
An AI that manages your to-do list, calendar, and reminders in one place—no more app overload.
MCPs make AI timely, relevant, and actionable. And for someone like me who loves to try new tech, this feels like an open invitation to start building. What would you build first?
Because if the last few years were about learning to talk to AI, the next ones will be about learning to work with it.
What are Model Context Protocols (MCP) in simple terms?
MCPs are like a universal translator and secure access pass for AI models. They provide a standardized way for an AI to connect to external tools, apps, and databases (like your calendar or a fitness app) to retrieve live, relevant information before it gives you an answer or performs a task.
Why are MCP important for the future of AI?
They allow AI to evolve from being a simple conversationalist with static knowledge to an actionable teammate that can interact with the real world. This means AI can perform multi-step tasks, provide personalized advice based on your current data, and automate workflows, making it much more useful in daily life and work.
What is a real-world example of how MCP work?
Imagine asking your AI to help plan your marathon training. Using MCPs, it could connect to your calendar to see your free time, your fitness app for your recent performance data, and a weather API. It would then use all this live context to give you a personalized, actionable training plan for the week ahead.
What are the biggest challenges facing MCP today?
The main challenges are security (ensuring AIs have the right permissions and nothing more), adoption (getting more apps and services to support these protocols), and latency (making sure the process of connecting to multiple sources is fast and efficient).
What is the key difference between a normal chatbot and an AI using MCP?
A normal chatbot operates on a fixed set of data it was trained on; its knowledge is locked in the past. An AI using MCPs can access live, real-time information from your personal apps and services, allowing it to provide answers and take actions that are timely, relevant, and specific to your current situation.
Comments