2026-03-22 17:24 Tags:Technical Literacy

https://youtu.be/7j1t3UZA1TY?si=x4iersxriU7AXcNu

---

Model Context Protocol (MCP) vs APIs

1. Why MCP exists

For large language models (LLMs) to be truly useful, they must:

• Access external data (documents, databases, knowledge bases)

• Use tools (APIs, services, actions)

Traditionally, this was done using APIs.

However, APIs were not designed specifically for AI agents.

→ This led to the introduction of Model Context Protocol (MCP) in late 2024 by Anthropic.

---

2. What is MCP

Definition:

MCP is an open standard protocol that standardizes how applications provide context and tools to LLMs.

---

3. Core Analogy (important)

MCP can be understood as:

USB-C for AI systems

Mapping:

Real World	MCP Equivalent
Laptop	MCP Host
USB-C Port	MCP Protocol
External devices (monitor, disk, power)	MCP Servers

Key idea:

• Different tools/services can connect using the same standard

• It does not matter who built them

• Everything works together through a unified interface

---

4. MCP Architecture

MCP follows a client–server model.

Components:

• MCP Host

  • Runs the AI application

• MCP Client

  • Opens a session using MCP protocol (JSON-RPC 2.0)

  • Communicates with servers

• MCP Server

  • Exposes capabilities (tools, data, prompts)

---

Flow:

LLM / Agent
   ↓
MCP Client
   ↓ (JSON-RPC)
MCP Server
   ↓
External systems (DB, APIs, services)

---

5. What MCP Enables

MCP solves two core needs of AI agents:

1. Context Retrieval

Access external data:

• documents

• database records

• knowledge bases

---

2. Tool Usage

Execute actions:

• web search

• API calls

• calculations

• system operations

---

6. MCP Primitives (Core Concepts)

MCP servers expose three main primitives:

---

6.1 Tools

Definition:
Discrete actions the AI can execute

Examples:

• get_weather

• create_event

Characteristics:

• Defined with name, description, input/output schema

• Executed by server when invoked

---

6.2 Resources

Definition:
Read-only data provided by the server

Examples:

• text files

• database schema

• documents

Used for:

• providing context to the model

---

6.3 Prompt Templates

Definition:
Predefined prompt structures

Purpose:

• guide model behavior

• standardize interactions

---

Important Insight

• Not all servers implement all primitives

• Many focus primarily on tools

---

7. Dynamic Discovery (Key Advantage)

MCP supports:

Runtime capability discovery

Meaning:

• Agent can ask:

  “What can you do?”

• Server returns:

  • available tools

  • resources

  • prompts

---

Implication:

• No need to hardcode integrations

• No redeployment required when capabilities change

• Agents adapt dynamically

---

8. What is an API

Definition:

An API (Application Programming Interface) defines how one system can request data or services from another.

---

Key Characteristics:

• Defines request/response structure

• Abstracts internal implementation

• Enables system integration

---

Example Use Case:

• E-commerce → payment API

• AI → web search API

---

9. REST API (Most Common Type)

REST APIs operate over HTTP.

Common Methods:

Method	Purpose
GET	Retrieve data
POST	Create data
PUT	Update data
DELETE	Remove data

---

Example:

GET /books/123
POST /loans

Responses are typically in JSON format.

---

10. Similarities Between MCP and APIs

Both:

• Use client–server architecture

• Provide abstraction layers

• Hide internal implementation details

• Enable system integration

---

Comparison Flow:

MCP	REST API
Client → MCP Server	Client → API Server
tools/call	HTTP request
structured response	JSON response

---

11. Key Differences

11.1 Purpose

MCP	API
Designed for AI agents	General-purpose
Focus on context + tools	Focus on data/services

---

11.2 Dynamic Discovery

MCP	API
Built-in capability discovery	Usually none
Agent adapts automatically	Requires manual updates

---

11.3 Standardization

MCP	API
Unified interface	Each API is different
Same structure across servers	Different endpoints, formats

---

Key Insight:

MCP = “build once, integrate many”
API = “integrate separately each time”

---

12. Relationship Between MCP and APIs

Very important:

MCP does NOT replace APIs
MCP builds on top of APIs

---

How they work together:

• MCP Server = wrapper layer

• Internally calls traditional APIs

---

Example:

MCP GitHub Server:

• Exposes:

  • repository/list (MCP tool)

• Internally:

  • Calls GitHub REST API

---

Conclusion:

MCP and APIs are layers in the same system

AI Agent
   ↓
MCP (standard interface)
   ↓
API (actual execution)
   ↓
External service

---

13. Why MCP Matters

MCP enables:

• standardized integration for AI agents

• dynamic tool usage

• reduced engineering overhead

• scalable system design

---

Real-world MCP integrations:

• file systems

• Google Maps

• Docker

• Spotify

• enterprise data systems

---

14. Key Takeaways

1. MCP standardizes how AI interacts with tools and data

2. It introduces dynamic discovery and unified interfaces

3. It simplifies multi-system integration

4. It works on top of existing APIs, not instead of them

---

One-line summary

MCP is a standardized interface layer that allows AI agents to discover, access, and use tools and data dynamically, often by leveraging APIs underneath.