MCP in plain English
MCP made simple
One shared way for AI apps to use tools and data
MCP replaces custom wiring with one standard connection.
The Problem
AI needs access.
Without a shared protocol, every app builds its own connections.
No access
The model cannot use files or data until it is connected.
Duplicate work
Teams rebuild the same integrations over and over.
Hard to maintain
When APIs change, custom connectors break in different places.
MCP gives AI one standard way to connect.
one standard, everywhere
The Journey
How MCP emerged
From no access to one shared standard
Just the model
Early models could answer, but not reach outside systems.
Custom integrations
Every app added its own connector for every service.
Model Context Protocol
One protocol that lets AI apps connect to compatible servers.
the standardThe Definition
What MCP is
An open standard anyone can read and build on.
Live context, delivered when the user asks.
Vendor-neutral across apps and stacks.
Think USB-C for AI tools: one plug, many devices.
FIG. 1 — one hub, many connections
The Motivation
Why MCP exists
Without MCP, integrations become a web. With MCP, they become a hub.
Select an application in either panel
Without MCP
FIG. 3 — the meshEach app talks to every service on its own.
- Direct integrations
- 9
- Custom SDKs
- 9
- Maintenance load
- High
With MCP
FIG. 4 — the hubEach app connects once. Each service is described once.
- Connections
- 6
- Protocol
- JSON-RPC 2.0
- Codebase footprint
- −65%
MCP is the standard layer for AI connectivity.
One protocol, defined once, reused everywhere.
The Demonstration
How it works
Toggle between the two setups to see the difference.
3 apps × 3 services = 9 integrations
Integrations
9
Maintenance load
High
Protocol overhead
N/A — custom SDKs
Select an application or service to trace a request.
Every app builds its own connector for every service.
The Architecture
How it fits
One request moves through five pieces.
FIG. 2 — the request path
User
You, asking the question
Host
The AI app you are using
MCP Client
The connector inside the app
MCP Server
Shows tools, data, and prompts
Protocol layer
External Service
Gmail, GitHub, databases, and more
Host
The AI app the user talks to, like Claude Desktop, an IDE, or a web app.
MCP Client
Lives inside the host and keeps one connection to one MCP server.
MCP Server
The middle layer that gives the client access to tools, data, and prompts.
The Building Blocks
What MCP shares
Resources
Things the AI can read.
Tools
Things the AI can do.
Prompts
Reusable instructions from servers.
The Payoff
Why it helps
You now know what MCP is, why it matters, and why the mesh gets messy.
Open the specFIG. 5 — next steps