Scribd
Upload
9 views3 pages

Understanding LLMs, Agents, and MCPs

The document explains key generative AI terminologies: LLM (Large Language Model), Agent, and MCP (Model Context Protocol). LLMs generate human-like text but cannot perform real-world actions, while Agents execute tasks based on LLM instructions, and MCP connects LLMs to external tools for practical tasks. Examples illustrate how these components interact in scenarios like browser automation and database operations.

Uploaded by

abhimanyu kumar
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
9 views3 pages

Understanding LLMs, Agents, and MCPs

The document explains key generative AI terminologies: LLM (Large Language Model), Agent, and MCP (Model Context Protocol). LLMs generate human-like text but cannot perform real-world actions, while Agents execute tasks based on LLM instructions, and MCP connects LLMs to external tools for practical tasks. Examples illustrate how these components interact in scenarios like browser automation and database operations.

Uploaded by

abhimanyu kumar
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Generative AI Terminologies: LLM, Agent, and MCP

LLM (Large Language Model):


An LLM is an advanced AI system trained to understand and generate human-like text.
It can read, write, and respond to questions in natural language.

Popular LLMs
• ChatGPT

• Google Gemini

• Claude

• DeepSeek

What LLMs Can Do


• Answer questions and explain topics

• Generate code or scripts

• Write emails, blogs, and documents

• Chat naturally with users

Example:
You can ask ChatGPT:

“Write a login test case in Java using Selenium.”


It will generate the code for you — but it won’t run the test itself.

Limitations of LLMs
LLMs are good at thinking and writing, but they can’t take real actions in the outside world.

They cannot:

• Open a browser or click buttons

• Connect to a database

• Call APIs to get live data

They can only generate the instructions or code, not execute them.

Agent:
An Agent is a system that takes help from an LLM and performs real-world actions using external
tools.

Think of it as an assistant that actually does what the LLM says.

[Link] [Link]
Example:
If the LLM says “Run this Playwright test,”
the Agent will execute that test in a real browser.

So:

• The LLM thinks and plans.

• The Agent executes and performs.

MCP (Model Context Protocol):


MCP is a framework that connects LLMs to real-world tools like browsers, databases, and APIs.
It acts as a bridge that allows LLMs to send commands to tools and get back the results.

What MCP Can Do


1. Browser Automation

o Open a website, fill forms, and click buttons

o Example: Selenium MCP or Playwright MCP

2. Database Operations

o Connect to MySQL or PostgreSQL and run queries

o Example: MySQL MCP or PostgreSQL MCP

3. API Requests

o Send or receive real-time API responses

o Example: API MCP for calling web services

So, MCP helps the LLM and Agent work together to perform live, practical tasks.

Playwright MCP:
Playwright MCP connects LLMs to browsers using the Playwright automation tool (similar to
Selenium).

Key Features
• Allows LLMs to control browsers (open sites, fill forms, click elements)

• Understands the web page structure using the Accessibility Tree (not images)

• Supports browser commands such as:

o click()

o fill('user@[Link]')

[Link] [Link]
Example
You tell ChatGPT:

“Search for ‘Selenium vs Playwright’ and click the first result.”

Behind the scenes:

• ChatGPT (LLM) sends this instruction to Playwright MCP.

• Playwright MCP opens a browser and performs the task automatically.

Summary
Term In Simple Words Example Scenario

LLM The brain that understands and generates GPT-5 explaining “What is API Testing?”
text.

Agent The assistant that uses the brain to perform Agent running a Playwright test based on
actions. ChatGPT’s instructions.

MCP The bridge that connects the brain and ChatGPT using MCP to open files or run
assistant to external tools. code in PyCharm.

[Link] [Link]

You might also like