MCP — the protocol
What MCP is, how it works, and how it connects Claude to everything
The plumbing that makes agents possible
MCP (Model Context Protocol) is an open standard developed by Anthropic that defines how AI models connect to external tools and data sources. Think of it as the USB-C of AI integrations — a universal connector that works the same way regardless of what you’re plugging in.
The three parts of MCP
MCP Host (the Claude client)
The application where Claude runs — Claude.ai, Claude Desktop, Claude Code, or a custom app. The host manages connections to MCP servers.
MCP Server
A process that exposes tools (callable functions) and resources (readable data) to Claude. An MCP server for Gmail exposes tools like list_emails, read_email, create_draft, send_email.
MCP Client (built into the host)
The component that manages the connection between the host and servers, passes tool calls from Claude to the right server, and returns results.
What MCP servers expose
| Concept | What it is | Example |
|---|---|---|
| Tools | Functions Claude can call | search_emails(query, date_range) |
| Resources | Data Claude can read | Your calendar for the next 7 days |
| Prompts | Reusable prompt templates | “Morning brief” prompt |
MCP in Claude.ai
In your Claude.ai account, you can connect MCP servers for Gmail, Google Calendar, Google Drive, and many third-party integrations. When these are connected, Claude can take actions in those systems during your conversations — not just talk about them.
Why it matters for agency work: MCP turns Claude from a writing assistant into an operator. Instead of Claude telling you “you should check your calendar,” it can check your calendar and tell you what it found. That’s the shift from advice to action.