Model Context Protocol · Complete Explanation
MCP — The Standard That Connects AI
to Tools & Data
The Model Context Protocol has quietly become the standard way AI connects to everything — your files, your databases, your tools. This deck breaks down what MCP actually is, how it works under the hood, where it sits inside an AI agent, and how it compares to function calling, APIs, and plugins.
Source: youtube.com/watch?v=_fzpnqt39jQ · Channel: Aishwarya Srinivasan (youtube.com/@aishwaryasrinivasan)
Agenda
What We'll Cover
Understanding MCP
- Why MCP exists — the drawer of chargers
- What it is — the USB-C for AI
- Host, client & server — the three pieces
- Tools, resources & prompts — what a server gives
- Where MCP sits inside an AI agent
- MCP vs function calling, APIs & plugins
Using & Building with MCP
- Finding servers — 10,000+ and counting
- Local vs remote servers, and how you use one
- The security reality no tutorial mentions
- The practitioner's safety checklist
- Building your first server with FastMCP
- Key takeaways
Why MCP Exists
The Drawer of Chargers Problem
An AI application's model needs to talk to your database, your CRM, your file system, maybe even Slack. Every one of those connections is a custom integration: you write it, you test it, you maintain it.
Point-to-point by default
Ten apps × a hundred tools ≈ a thousand bespoke integrations.
Connectors wired by hand that only work for the exact pair they were built for — each one brittle.
One cable per device
The drawer of chargers we all had before USB-C.
The AI ecosystem has hit the same wall: many models, many tools, many data sources, all wired together by hand.
What MCP Is
The USB-C for AI
An open standard that gives AI models and the tools and data around them one common language for connecting — you build a connector once, and any MCP-compatible application can use it.
Build once
Claude, ChatGPT, Cursor, or your own custom agent — no rewrites and no special casing per platform.
The default way
MCP became the standard way an AI application connects to the outside world.
One agreed plug
The industry essentially agreed on the shape of the plug — and it is not a one-company thing anymore.
How MCP Works · The Three Pieces
Host, Client, Server
There are exactly three pieces — and the USB-C analogy maps onto them almost perfectly. Hold these in your head and every MCP diagram you have squinted at makes sense.
Host
The AI application — like your phone.
Claude Desktop, Cursor, VS Code, ChatGPT, and so on. The host is the app you actually use.
MCP client
The USB-C port inside the host.
Speaks the protocol and manages the connection. From your point of view it is invisible plumbing; to the protocol it is the endpoint doing the talking.
MCP server
The accessory you plug in.
A small program that wraps tools and data sources — database, calendar, GitHub — in the standard MCP format. It can literally be a hundred lines of Python on your laptop.
The client asks what the server has; the server advertises its offerings — nothing is hardcoded anywhere. That discovery step is on the next slide.
What a Server Gives
Tools, Resources & Prompts
A server gives the model three kinds of things. The clearest way to see the difference: imagine onboarding a new employee.
Tools
What the model can do.
Actions — like giving the new hire software access. A filesystem server exposes read and write operations as tools.
Resources
What the model can read.
Data — like handing the new hire the company wiki. The model can open files as resources.
Prompts
Reusable templates.
For doing tasks well — like standard operating procedures for common workflows.
Discovered at runtime — why it feels like magic
When a client connects, it simply asks "What do you have?" The server advertises its tools, resources, and prompts — descriptions and typed inputs included. Nothing is hardcoded into the client or the model. Add a server, and the application instantly knows what it offers and how to call it.
Where MCP Sits
MCP Alone Is Not an Agent
When people say "AI agent," what is really running is a harness — a loop around the model. The model does the reasoning; the harness handles everything else. MCP is the access layer inside that system.
Model — the brain
Does the reasoning and decides what the agent needs next.
Harness — the workstation & workflow
Manages memory, tracks state, decides when to go around the loop again, and executes actions in the real world.
MCP — the company badge
The standardized access layer into every system the agent is allowed to touch — no bespoke key for each door.
Where MCP Sits · The Loop
Model Decides, Harness Executes, MCP Connects
The loop in motion — repeated until the task is complete:
MCP vs Function Calling & APIs
Complementary Layers, Not Replacements
| Layer | What it is | The framing |
|---|---|---|
| Function calling | A capability of the model: it decides a tool is needed and produces a structured request with the right arguments | "Should I make a call, and what should I say?" |
| MCP | The protocol that carries that request to the tool and brings the result back | "How does that call actually reach the other end?" |
| Plain APIs | Point-to-point connections to one service — the API still does all the work | MCP is the standard layer on top, giving every model the same way to find and call it |
You need both: function calling is deciding to call somebody and dialing the number; MCP is the telephone network that lets any phone reach any other phone.
MCP vs Plugins
Why the Plugin Era Quietly Ended
2023 · ChatGPT plugins
Proprietary: one cable per brand.
You built a plugin for one platform — and if you wanted the same capability anywhere else, you had to build it again.
MCP flips the model
One open protocol, one port everywhere.
Every AI platform implements the same port. Build your server once and it works on Claude, ChatGPT, Cursor, VS Code, and whatever agent framework you run.
The outcome
Finding MCP Servers
Search Before You Build
There are over 10,000 public MCP servers out there at last count — discovery is genuinely the easy part. Three places to look:
Official MCP registry
The canonical directory, where verified publishers increasingly list their servers.
GitHub servers repo
The Model Context Protocol servers repository — reference servers including filesystem, fetch, and memory — one of the most-starred repos in the space.
Community directories
Pulse MCP and similar — browse by category: a Notion server, a Postgres server, a GitHub server.
Local & Remote Servers
One Fork in the Road
Local server
Runs on your machine, over stdio.
Standard input and output between processes. It is what you want for local files and experimentation.
Remote server
Hosted elsewhere, over HTTP.
Typically authenticated with OAuth — the same way you sign in to any app with your Google account. Remote is how most companies expose their products.
Using a server turns out to be anticlimactic. Recommended start: go local with something low-stakes, like a filesystem server, and build your intuition before scaling to many servers.
The Security Reality
Only 13% Make the Trust Bar
An MCP server is code that gets a direct line into your AI's context — and often into your real accounts and data. This is the part almost no MCP tutorial talks about, and it deserves to be taken seriously.
What "trust thresholds" cover
Documentation, maintenance, and reliability — as measured by independent assessments. And the incidents above actually happened — they are not hypotheticals.
Using MCP Safely
The Practitioner's Checklist
Individual practice
- Prefer official servers from verified publishers — or read the code before you run it; server code is usually small enough that you actually can.
- Least privilege, always. If a server only needs to read data, give it read-only credentials.
- Keep a human in the loop for everything sensitive or irreversible — writes, deletes, sending messages.
Doing this at team scale
Put your MCP servers behind a central gateway or an internal registry.
- Control exactly which servers your agents can reach
- Default everything to scoped read-only credentials
- Put sensitive operations behind human approval
- Handles thousands of tool calls a day — without ever handing the agent the keys to everything
Build Your First MCP Server
A Fifteen-Minute Project
FastMCP lives in the official Python SDK: write a normal Python function with type hints and a docstring, add a decorator, and the SDK turns it into a fully described MCP tool — no protocol plumbing by hand. For TypeScript developers, the official SDK is just as solid.
Validate in the Inspector, connect it to your client — and don't overbuild: most people overbuild their first server, so start with one well-tested tool over stdio.
Key Takeaways
The Complete Checklist
The End
It Is the Plumbing That Wins
Teams winning with AI right now are exactly the ones doing that plumbing: giving their models a standardized way to actually touch real-world data.
Source: "MCP Complete Explanation" — Aishwarya Srinivasan · youtube.com/watch?v=_fzpnqt39jQ · youtube.com/@aishwaryasrinivasan