MCP, the Model Context Protocol, is an open standard that lets AI apps connect to outside tools and data in one shared way. A system is wrapped once as an MCP server, and any MCP-ready app can then use it.

Is MCP the same as an API?

Not exactly. An API is one service’s own way of being called. MCP is a common layer on top, so AI apps can discover and use many different services the same way. An MCP server often calls the service’s API behind the scenes.

Do I need to code to use MCP?

Not to use it. Many AI apps let you add an MCP server from their settings. Building your own server does need some coding, and official kits exist for common languages.

Module 5 · AI that takes action: agents · Lesson 24 of 32

MCP: one plug for every tool

9 minFree GenAI course · BeginnerLesson 5.3By Saket Bhatnagar

Researched and drafted with AI, reviewed and edited by Saket Bhatnagar.

In short

MCP, the Model Context Protocol, is an open standard for connecting AI apps to tools and data.
A tool is wrapped once as an MCP server, and then any MCP-ready AI app can use it, like one USB-C charger for every phone.
MCP makes connecting simpler, not automatically safer: you still choose which servers to trust and what needs your approval.

Soon

Video lesson · coming soon

What is MCP? The Model Context Protocol in 3 minutes

Lala, your AI mentor, will walk you through this lesson in a short animated video. Until then, everything is here to read and try.

01 The charger drawer problem
02 Three apps, four tools, twelve wires
03 MCP is UPI for AI tools
04 Host, client, server
05 Flip: before and with MCP
06 A plug, not a safety stamp
07 Try the cards, then see AI images

Open the drawer in any Indian home and you will find a tangle of chargers. The thin round pin for an old phone, the micro-USB, the one that only fits Papa’s tablet. Every new device meant hunting for the right cable.

Then USB-C arrived. One kind of plug, and the same cable charges your phone, your laptop and your earbuds. AI apps had a charger drawer problem too, until a shared plug for tools came along. That plug is MCP.

The problem: every app wired to every tool

In the last two lessons, a model asked for tools and an agent used them in a loop. But someone has to write the code that connects each AI app to each tool: the calendar, the email, the company database, the code repository.

Say your college builds three AI apps: a chat helper, a coding assistant and a library bot. Each needs four tools. Wired one by one, that is three times four, twelve separate connectors. Each speaks its own format, and each breaks when a tool changes.

With one shared standard, each app learns the standard once and each tool is wrapped once. That is three plus four, seven pieces. A fifth tool costs one new piece, not three. This is the whole reason MCP exists.

Three apps, three tools, one standard

Drag to explore

Each AI app speaks MCP once and each tool is wrapped as one MCP server. Any app can now use any tool, without a custom connector for every pair.

MCP was first published as an open standard in late 2024. Through 2025, most major AI chat apps, coding tools and code editors added support, and thousands of MCP servers appeared. Today the project is looked after by an open foundation under the Linux Foundation, so no single company owns it.

Try it yourselfBefore MCP, and with MCPRead each “before” story, then flip the card to see how the same job looks once both sides speak MCP.

Host, client and server: the three roles

MCP uses three words you will hear often. They sound technical, but they map neatly onto what you already know from tool calling.

An MCP server can run on your own laptop, for example to read files in one folder. Or it can live on the internet, run by the company behind a service. Either way, the host talks to it in the same format.

What an MCP server can offer, and who decides when it is used
What it offers	What it is	Who decides when	Example
Tools	Actions the model can ask to run	The model asks; the app checks	create_event, search_orders
Resources	Data the app can attach as context	The app	A project README, a table’s layout
Prompts	Ready-made instructions a person can pick	You	A “summarise this meeting” template

Here is the key link to the last lessons. MCP does not replace tool calling. The host reads the tool list from MCP servers and offers it to the model as its tool menu. The model asks for a tool exactly as before. MCP only standardises how those tools are found and called.

json

{
  "name": "find_free_slots",
  "title": "Find free calendar slots",
  "description": "List free 30-minute slots in the user's calendar on a given date.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "date": { "type": "string", "description": "Date as YYYY-MM-DD" }
    },
    "required": ["date"]
  },
  "annotations": { "readOnlyHint": true }
}

Notice it looks a lot like the weather tool from the tool calling lesson. The readOnlyHint tells the host this tool only reads, so it is a good candidate to run without asking you every time.

What MCP does not do, and how to stay safe

A standard plug does not make every charger safe. A cheap, badly made charger can still damage your phone. MCP is the same: it makes connecting simpler, but a server you plug in can be careless or even harmful.

Install servers only from sources you trust, the way you would only install a well-known app.
Give each server the smallest access it needs. A calendar server that only reads cannot delete your meetings.
Keep approval on for actions. Sending, paying, deleting and posting should wait for your yes.
Remember that tool descriptions and tool results are text the model reads. A bad server could hide instructions in them.
Watch the menu size. Every tool description costs tokens in the context window, so connect what you use, not everything you can find.

Check yourself

Five AI apps each need the same six tools. Why does MCP help?

Pick an answer to see why it is right or wrong.

Check yourself

In MCP, which part is the AI app you actually type into?

Pick an answer to see why it is right or wrong.

Check yourself

A friend says, “It is an MCP server, so it must be safe to install.” What do you tell them?

Pick an answer to see why it is right or wrong.

Remember these three

MCP is an open standard plug between AI apps and tools, so each tool is wrapped once and works everywhere.
The host is your AI app, each client links to one server, and each server wraps one system’s tools and data.
MCP standardises connecting, not trust: pick servers carefully, limit access and keep human approval for actions.

Questions people ask

What is MCP in AI?: MCP, the Model Context Protocol, is an open standard that lets AI apps connect to outside tools and data in one shared way. A system is wrapped once as an MCP server, and any MCP-ready app can then use it.
Is MCP the same as an API?: Not exactly. An API is one service’s own way of being called. MCP is a common layer on top, so AI apps can discover and use many different services the same way. An MCP server often calls the service’s API behind the scenes.
Do I need to code to use MCP?: Not to use it. Many AI apps let you add an MCP server from their settings. Building your own server does need some coding, and official kits exist for common languages.
Who owns MCP?: MCP is an open standard. It was first published in late 2024 and is now looked after by an open foundation under the Linux Foundation, with the specification public at modelcontextprotocol.io.

Sources and further reading

Model Context Protocol: official site and specification