What is Model Context Protocol?

DeepLearning
AI
Agents
LLMs
A blog post about the Model Context Protocol
Published

April 3, 2023

Model Context Protocol

Everybody is talking about the Model Context Protocol (MCP) these days. But what is it?
Model context protocol is a standardized way of formatting prompts and responses between humans and AI language models. It was developed by Anthropic and is now supported and used by many other LLM providers. The protocol helps establish clear boundaries and roles in human-AI interactions by using specific markers like “Human:” and “Assistant:” to indicate who is speaking. This structured format makes it easier for language models to understand the conversation flow and maintain consistent behavior.

What is MCP?

Model Context protocol MCP is standard protocol which is used to connects different AI agents to external tools and services.

https://www.anthropic.com/news/model-context-protocol

API vs MCP

Generally external services are connected using API which differs from providers to providers. It’s a pain to manage differnet API follow different SDKs, documentation, auth setup and maintaining them.

MCP is open standard which is being quickly adopted by many product developers, companies and organizations.

MCP Features

  • Single Protocol: Universal connector to integrate any service or tool.
  • Dynamic Discovery: Automatically discover and connect to new tools. (similar to package managers)
  • Two-way Communication: Agents can send (trigger events) and receive messages to and from external tools at the same time.

MCP Components

MCP Host: Claude desktop or AI driven IDE which requires external tools to be connected to it. MCP Client: External tool or service which is connected to MCP server. MCP Servers: Lightweight servers which are used to connect MCP clients to MCP host. Local Data Source: Local files, databases, etc. Tool/Services: APIs of external tools and services.

Just checkout the list of MCP servers and integrations below. https://github.com/modelcontextprotocol/servers?tab=readme-ov-file

To use MCP in your projects, can follow the notes link below. Model Context Protocol Example