What are MCPs?

Understanding Model Context Protocol and why it matters

Understanding MCPs

MCPs are the UI for AI agents. Just like humans need buttons and forms, AI agents need MCPs to understand your tools.

The Problem

Your tools work great for humans and developers:

• Humans use buttons, forms, and visual interfaces
• Developers use APIs with documentation
• AI agents get confused without proper guidance

Without MCPs, AI is like a user on a webpage with 1000 random buttons scattered everywhere. They don’t know:

• Which button does what
• When to click which button
• What happens after clicking

What MCPs Provide

MCPs solve this by giving AI three things:

1. Tools (Actions)

What AI can DO with your system:

{
  "name": "send_email",
  "description": "Send an email to someone", 
  "parameters": {
    "to": "email address",
    "subject": "email subject",
    "body": "email content"
  }
}

2. Resources (Data)

What AI can ACCESS from your system:

{
  "uri": "user://contacts",
  "name": "User Contacts",
  "description": "List of user contact information"
}

3. Prompts (Templates)

What AI can USE as starting points:

{
  "name": "meeting_invite",
  "description": "Create a meeting invite email",
  "template": "Hi {name}, let's meet on {date} at {time}..."
}

Why This Matters

For AI Agents

• Clear guidance - Know exactly what tools are available
• Better decisions - Understand when to use each tool
• Fewer errors - Proper schemas prevent mistakes

For You

• Better results - AI uses your tools correctly
• Less confusion - AI doesn’t make random tool calls
• More adoption - People use tools that work well

Real Example

Without MCP:

User: "Send John a meeting invite"
AI: "I don't know how to send emails or who John is"

With MCP:

User: "Send John a meeting invite"
AI: "I'll check your contacts for John and send him a meeting invite"
   → Uses contact resource to find John
   → Uses email tool to send invite
   → Uses meeting template for content

MCP vs API

API	MCP
For developers	For AI agents
Static documentation	Dynamic discovery
Human interpretation	AI interpretation
Complex integration	Simple integration

The Art of Good MCPs

Building good MCPs is like designing good user interfaces:

❌ Bad UI

• 1000 buttons everywhere
• Unclear labels
• No organization
• Confusing workflows

✅ Good UI

• Clear, organized sections
• Descriptive labels
• Logical grouping
• Intuitive flow

❌ Bad MCP

• Too many similar tools
• Vague descriptions
• Complex schemas
• Overlapping functionality

✅ Good MCP

• Focused tool set
• Clear descriptions
• Simple schemas
• Distinct purposes

Next Steps

Best Practices

Learn how to design MCPs that AI loves

Testing Guide

Test your MCPs with real AI agents

Quick Start

Build your first MCP in 5 minutes

Command Line Interface

Use the CLI for local development

API Reference

Explore the LeanMCP API

Key Takeaway

MCPs are not just technical specifications - they’re the interface between your tools and AI intelligence.

The better your MCP, the better AI can help your users accomplish their goals.