Learn/L13 · Model Context Protocol (MCP)

What is Model Context Protocol (MCP)?

MCP is a standard way for AI assistants to use tools and access context.

Difficulty 4/5

Mental model

Plain English

An AI like Claude or Cursor is just a model. MCP lets it talk to a 'tool server' that exposes things it can do — read your GitHub, query your database, list your calendar. The model says 'I want to call tool X with these args'; the MCP server executes it; the result comes back into the conversation.

Analogy

MCP is USB for AI assistants. Any tool can plug in if it speaks the protocol.

Jessica's wedding portal

An MCP server exposes 'list_tasks', 'create_task', 'list_vendors'. The AI can prep a checklist for next week's wedding.

Agentify CRM

An MCP server exposes 'list_conversations', 'send_message', 'assign_agent'. Claude can run agent workflows on Agentify data.

Go deeper

USB-C for AI assistants.

Common mistakes

Giving an MCP server too much access. Scope it tightly.
Trusting model output to call the right tool. Validate args server-side.
Forgetting that MCP servers run code on your machine. Audit what they do.

AI-agent trap

Agents will happily install untrusted MCP servers. Treat them like browser extensions — only install from sources you trust.

Quick check

What does an MCP server expose to the AI?

Summary

MCP defines: tools (functions the AI can call), resources (read-only data), and prompts (templated workflows). The client (Claude Desktop, Cursor) connects to one or more MCP servers over stdio or HTTP.

Confidence: