Скачать или смотреть MCP Course #2 - MCP Basics - Quick Overview

End-to-end MCP Course with full code.

Follow @theailanguage on X
https://twitter.com/intent/user?scree...

Subscribe to youtube
   / @theailanguage  

Happy building! ☺️
---------------------------------------------------------------------------------------------
Udemy Course (get completion certificate, practice questions, Q&A) https://www.udemy.com/course/modelcon...
---------------------------------------------------------------------------------------------

MCP by @AnthropicAI

1/ Purpose: protocol to connect LLMs to external data sources and tools in a universal way

2/ Client-Server Model: MCP client connects to MCP servers
Client: an LLM-powered app like Claude Desktop
Server: programs that expose tools, resources, or prompts

3/ MCP Servers provide three primitives -
Prompts: reusable templates with instructions or context to guide the LLM's behaviour
Resources: access to external data & files for the LLM to retrieve information
Tools: functions or APIs for the LLM to invoke to perform actions

4/ MCP Clients provide two primitives -
Roots: interfaces or entry points through which the client connects to MCP servers
Sampling: mechanism to allow the server to request the client to generate text completions based on specific inputs or contexts

5/ Bidirectional: with these primitives, MCP supports two-way communication
For instance -
A server can request the LLM to generate completions (Sampling primitive)
A client hosting the LLM can request a tool execution

6/ Pre-built MCP servers exist for GitHub, Google Drive, Postgres, etc. Early Stage, success depends on wider adoption

7/ Example - MCP server to connect an LLM to a local SQLite database, GitHub repo, and Slack workspace. LLM can query database, fetch code context, and post updates—all within a single protocol