Model Context Protocol (MCP) is a standardized way for AI models to interact with external systems, tools, data sources, and contextual information.
Think of MCP as a universal connector between an AI model and the resources it needs to perform useful work.
Without MCP:
User → LLM → Response
With MCP:
User
↓
LLM
↓
MCP
├─ Files
├─ Databases
├─ APIs
├─ Vector DBs
├─ Enterprise Applications
└─ Tools
The primary goal of MCP is to help AI systems:
Access external knowledge
Use tools and APIs
Maintain context
Interact with enterprise systems
Execute actions safely and consistently
Analogy:
MCP is to AI what USB is to hardware: a standard way to connect different components without custom integrations for every new system.
Why Do We Need MCP?
A standalone LLM has limitations:
❌ Doesn't know your company's latest data
❌ Cannot directly query databases
❌ Cannot call APIs by itself
❌ Cannot access internal documents
❌ Cannot execute business workflows
MCP solves these limitations by enabling structured communication between the model and external resources.
Core Components of MCP
A typical MCP architecture consists of the following components:
1. Client
The client is the AI application that initiates requests.
Examples:
Chatbot
Copilot
Web application
Agentic workflow
User
↓
MCP Client
Responsibilities:
Send user requests
Request tools/resources
Receive responses
2. Host (LLM Application)
The host is the application running the AI model.
Examples:
ChatGPT-like application
Microsoft Copilot
Enterprise assistant
Custom AI agent
Responsibilities:
User
↓
Host Application
↓
MCP
3. MCP Server
The MCP Server exposes capabilities to the AI model.
Examples:
Database access
CRM system
HR application
File storage
Knowledge repository
MCP Server
├─ Database
├─ File System
├─ API
└─ Business Application
Responsibilities:
Expose resources
Expose tools
Return data
Execute operations
4. Resources
Resources provide data to the model.
Examples:
PDF documents
Text files
SharePoint content
Database records
Knowledge articles
Example:
Employee Handbook
Leave Policy
IT Support Guide
The model can retrieve these resources through MCP.
5. Tools
Tools allow the AI to perform actions.
Examples:
Send email
Create ticket
Query database
Update CRM record
Fetch employee details
Example:
get_employee_details(employee_id)
Instead of answering from memory, the model can call the tool.
6. Prompts
MCP can expose reusable prompts.
Example:
Summarize this document
or
Generate a project status report
These prompts standardize common workflows.
MCP Workflow (Step-by-Step)
Let's use a simple scenario:
User asks:
Show employee details for employee 101
Step 1: User Sends Request
User
↓
Host Application
Example:
Show employee details for employee 101
Step 2: LLM Understands Intent
The LLM analyzes the question.
It determines:
Step 3: Discover Available Tools
Through MCP, the model sees:
Tool:
get_employee_details()
Step 4: Tool Call Generated
The model generates:
{
"tool": "get_employee_details",
"employee_id": "101"
}
Step 5: MCP Server Executes Tool
LLM
↓
MCP Server
↓
HR Database
Example:
SELECT * FROM employees
WHERE employee_id = 101
Step 6: Server Returns Results
{
"name": "John Doe",
"title": "Engineer"
}
Step 7: LLM Generates Human-Friendly Response
The model receives the result and responds:
Employee 101 is John Doe and works as an Engineer.
Step 8: Response Sent to User
User
↑
Response
Process completed.
MCP Workflow Diagram
User
↓
Host Application
↓
LLM
↓
MCP Client
↓
MCP Server
↓
Tool / Resource
↓
Result
↓
LLM
↓
Response
↓
User
MCP and RAG Relationship
Many people ask:
Is MCP the same as RAG?
No.
RAG
Focuses on:
Retrieve information
+
Generate answer
Example:
Question
↓
Vector Search
↓
Documents
↓
LLM
↓
Answer
MCP
Focuses on:
Tools
Resources
Prompts
RAG
Memory
Context
Example:
Question
↓
MCP
├─ RAG
├─ Tools
├─ APIs
├─ Files
└─ Databases
↓
LLM
RAG is often one capability within an MCP-enabled architecture.
Real-World Example
Imagine an HR Copilot.
User asks:
How many leave days do I have?
MCP workflow:
Output:
You currently have 12 leave days remaining.
The answer comes from the HR system, not from the model's training data.
Benefits of MCP
✅ Standardization
One protocol for multiple systems.
✅ Tool Integration
Easy access to APIs and applications.
✅ Security
Controlled access to enterprise data.
✅ Scalability
Add new tools without rewriting the AI application.
✅ Interoperability
Different AI models can use the same MCP server.
Summary
MCP (Model Context Protocol) is a standard protocol that enables AI models to securely access and interact with external tools, resources, databases, APIs, and enterprise systems.
Core Components
Client – initiates MCP requests
Host – AI application running the LLM
MCP Server – exposes capabilities
Resources – data sources
Tools – executable actions
Prompts – reusable workflows
Workflow
User Request
↓
LLM Understands Intent
↓
Discover MCP Tools/Resources
↓
Call Tool or Resource
↓
MCP Server Executes Request
↓
Return Results
↓
LLM Generates Final Answer
↓
User Receives Response
In one sentence:
MCP provides a standardized bridge between AI models and the external world, allowing them to retrieve information, use tools, and perform real business tasks.