Model Context Protocol
What is the Model Context Protocol?
The Model Context Protocol (MCP) is an open standard developed and open-sourced by Anthropic that enables AI models to connect securely to diverse data sources and tools through a unified interface.
The protocol handles data serialization, discovery and more.
"Enables models to call APIs"
"Empowering AI to be proactive and integrate seamlessly into workflows"

Key Innovation
MCP transforms AI integration from a many-to-many problem into a one-to-many solution through standardization.
USB-C for AI Applications
Just as USB-C created a universal standard that works across devices, MCP provides a universal protocol that works across AI hosts and data sources. One standard, infinite possibilities.
MCP - Model Context Protocol
The Model Context Protocol (MCP) facilitates secure and efficient interaction between AI applications and various data sources and resources. It establishes a standardized method for integration and data exchange.
The Integration Paradigm Shift
Before MCP
If you wanted Claude or ChatGPT to access your Google Drive, Slack, and PostgreSQL database, the AI company had to build specific integrations for each one.
With MCP
You build a single "Server" that speaks MCP. Any AI "Host" (like Claude Desktop, Cursor) can plug into that server and instantly understand how to read your data or perform actions.
  • Build once, use everywhere
  • Standard protocol reduces complexity
  • Any MCP-compatible host works immediately
  • Rapid ecosystem growth
Core MCP Architecture
MCP consists of three distinct components that work together to enable seamless AI-to-data communication.
MCP Host
The AI application itself (examples: Claude Desktop, Zed, Cursor). The Host is responsible for initiating the connection and orchestrating the overall interaction flow.
MCP Client
The communication bridge embedded inside the Host that handles protocol translation and message routing between the Host and Server components.
MCP Server
A lightweight program that exposes your data sources or tools through the MCP interface, translating between your systems and the standard protocol.
MCP Server endpoints
  • Capabilities ( Discovery )
  • Tools
  • Resources
  • Prompts
MCP tools often provide OpenAPI specs
  • Defines endpoints, parameters, authentication.
  • Enables models to understand API structure automatically
Resources: Passive Data Access
What are Resources?
Resources represent passive data sources that the AI can read from. The server declares available data, and the AI can request specific resources as needed.
Common Examples
  • Log files and system outputs
  • API endpoint responses
  • File system documents
  • Configuration data
Use Case
"Claude, analyze the error logs from the last hour and identify the most frequent failure patterns."
The AI requests the log resource, receives the data, and performs analysis without executing any actions.
Tools: Executable Actions
Tools are executable functions that the AI can invoke to perform actions in your systems. Unlike resources which are read-only, tools enable the AI to make changes and trigger operations.
Communication Tools
Send emails, Slack messages, or notifications
Data Operations
Execute SQL queries, update records, run migrations
Workflow Actions
Create Jira tickets, update project management systems
How Tools Work
When the AI decides to use a tool, it sends a JSON-RPC request to the server specifying the tool name and parameters. The tool executes the action and returns a structured result.
Example: "Claude, create a high-priority ticket for this bug with logs attached." The AI calls the createTicket tool with parsed parameters, the tool executes against your system, and returns the ticket ID and URL.
Prompts: Templated Workflows
Reusable Prompt Templates
Prompts are pre-written workflows or prompt templates stored on the server that users can invoke directly from the AI interface.
Rather than typing the same complex instructions repeatedly, you can create standardized prompts that automatically gather necessary context and format requests consistently.
Benefits
  • Consistency across team members
  • Reduced cognitive load
  • Best practices enforcement
  • Context automation
Code Review Prompt
Automatically grabs context from repository structure, recent commits, and coding standards to perform comprehensive reviews
Debug SQL Prompt
Gathers schema information, recent queries, and error logs to assist in troubleshooting database issues
Documentation Generator
Analyzes code structure and generates consistent API documentation following your team's format
Use case: Users select a prompt from a menu in the AI interface, and the prompt automatically executes with relevant context pre-loaded.
Communication Protocols
MCP uses JSON-RPC 2.0 for message formatting, but the transport layer—how those messages physically move between components—offers two main options depending on your deployment architecture.
STDIO (Standard Input/Output)
Most Common for Local Integrations
This is the default for local integrations such as code editors or the Claude Desktop app.
How it works: The Host launches the Server as a subprocess, and they communicate via stdin and stdout streams.
Advantages:
  • Very fast with minimal overhead
  • Easy to set up locally
  • Secure—no network ports opened
  • Ideal for development environments
SSE over HTTP
Best for Remote Deployments
Used when the AI Host runs on one machine and the Server runs on another, or for production deployments.
How it works: Standard HTTP POST for client-to-server messages, Server-Sent Events (SSE) for server-to-client updates.
Advantages:
  • Works across network boundaries
  • Supports distributed architectures
  • Standard web protocols
  • Scalable for production use
JSON-RPC in MCP
  • Some tools use JSON-RPC instead of REST
  • Remote Procedure Call using JSON messages
  • { "jsonrpc": "2.0", "method": "createrUser", "params": {userName:"gal", age:37, email:"[email protected]"}, "id": 1 } /mcp /POST
  • MCP models can invoke JSON-RPC endpoints just like REST endpoints.
  • Communication between Microservices
Key Differences: API vs MCP
Example
n8n
Node.js modelcontextprotocol