🔌

MCP (Model Context Protocol)

A universal standard for connecting AI assistants to external services. Think of it as installing apps on your phone, but for your AI.

Concept·7 sections·1 min read

01The Simple Explanation

Think of your phone

When you first buy a phone, it can make calls and send texts. Useful, but limited. Then you install apps. Instagram lets you share photos. Uber lets you order rides. Spotify lets you play music. Each app gives your phone a new ability it did not have before.

MCP works the same way for AI assistants like Claude.

🐙

Add GitHub MCP

Claude can create repositories, open pull requests, review code, and search issues.

📋

Add Notion MCP

Claude can read and edit your documents, search your workspace, and manage databases.

📧

Add Gmail MCP

Claude can read your emails, draft replies, and flag what is urgent.

✅

Add Todoist MCP

Claude can create tasks, organize projects, and track your to do list.

Each MCP connection gives Claude a new ability. The more you add, the more capable it becomes.

02Why MCP Exists

Before MCP, connecting an AI assistant to an external service required custom code for every single integration. If you wanted Claude to talk to five services, someone had to build five completely separate integrations. If a new service appeared, someone had to build integration number six from scratch.

Before MCP

Integration model

Custom code for every service. Five services means five separate integrations.

Analogy

Every device needs a different cable. One for monitor, another for phone, another for headphones.

New service added

Build a new integration from scratch. Weeks of work.

With MCP

Integration model

One universal standard. Build once, works with every compatible AI tool.

Analogy

USB C. One cable, any device. Everything just works together.

New service added

Create one MCP server. Works with Claude Code, Cursor, Codex CLI, and every other compatible tool.

Universal Standard

One protocol for unlimited connections. MCP is the USB C of AI.

03How It Works

Let us walk through a real example. You have Claude Code running with a Gmail MCP connected.

You type: "Do I have any urgent emails this morning?"

Step 1: Claude reads your message

It understands you want email information.

Step 2: Claude checks its available tools

It finds the Gmail MCP, which has tools for reading emails.

Step 3: Claude calls the Gmail MCP tool

It sends a request: "get recent emails."

Step 4: The MCP server translates the request

It converts Claude's request into something Google's email system understands.

Step 5: Google processes the request

Using your credentials, it retrieves your emails and sends them back.

Step 6: The MCP server passes data back

Claude receives the email data in a format it can read.

Step 7: Claude summarizes the results

It reviews the emails, identifies what looks urgent, and gives you a clean summary.

All of this happens in seconds. You asked a question in plain English and got a useful answer, without ever opening Gmail yourself.

04What Is an MCP Server?

Forget about data centers and blinking lights

The word "server" sounds intimidating. An MCP server is just a translator.

Imagine two people who speak different languages sitting in a room. One speaks English (Claude). The other speaks Spanish (Google, GitHub, or Notion). They cannot communicate directly. So you bring in a bilingual person who speaks both languages. That person listens to one side, translates, and passes the message to the other.

That bilingual person is the MCP server.

	Local (STDIO)	Remote (HTTP)
Location	Runs on your computer	Runs on the internet
How it starts	Claude Code launches it automatically	Claude Code connects via a URL
Analogy	A translator sitting in your office	A translator you call on the phone
Examples	Firecrawl, Yahoo Finance, Puppeteer	GitHub, Notion, Google Workspace

Both types work identically from your perspective. You ask Claude to do something, and it uses whichever MCP server has the right tools. You do not need to think about which type it is.

05Configuration

Think of a contact list

Each entry has a name and an address. When you want to reach someone, your phone looks them up and connects you. Claude Code has a similar contact list: a configuration file at ~/.claude.json. Each entry has the name of an MCP server and its "address" (how to connect to it).

# Example MCP configuration
{
  "mcpServers": {
    "github": {
      "type": "http",
      "url": "https://api.githubcopilot.com/mcp"
    },
    "notion": {
      "type": "http",
      "url": "https://mcp.notion.com/mcp"
    },
    "firecrawl": {
      "type": "stdio",
      "command": "npx",
      "args": ["firecrawl-mcp"]
    }
  }
}

➕Add an entry to give Claude a new ability

➖Remove an entry to take an ability away

🔧type says whether it is local (stdio) or remote (http)

06Available MCP Servers

	Connects To	What Claude Can Do
GitHub	Code repositories	Manage repos, create PRs, review code, search issues
Firecrawl	The web	Search the internet, scrape pages, extract data from websites
Notion	Documents and databases	Read pages, edit content, search your workspace
Todoist	Task management	Create tasks, manage projects, track progress
Google	Calendar, email, docs	Check schedule, find free time, manage events
Perplexity	AI research	Deep research with sourced, cited answers
Puppeteer	Web browsers	Control a browser, take screenshots, fill forms
Yahoo Finance	Stock market data	Get prices, financials, recommendations

Servers Documented

Server Types

Universal Standard

07Quick Recap

📱MCP is like installing apps on your phone. Each one gives your AI assistant a new ability.

🗣️MCP servers are translators between Claude and external services.

🔗One universal standard means any MCP server works with any compatible AI tool.

📇The config file is your contact list. Add an entry and Claude gains a new connection.

🏠Two types exist (local and remote), but they work the same from your perspective.

✨You do not need to be technical to benefit. Setup is usually one command or one config entry.