MCP — Connect Claude to Everything

Model Context Protocol: the architecture

MCP is an open standard that lets Claude connect to external tools and data sources through a simple server/client architecture. Here's what's actually happening:

1. An MCP server is a small program running on your computer. It exposes tools — functions Claude can call.
2. Claude Desktop acts as an MCP client. It discovers your installed servers when it starts.
3. When you ask Claude to read a file, Claude calls the read_file tool on the filesystem server. The server reads the file and returns the content.
4. Claude uses that content in its response — as if you had pasted the file in yourself.

The key insight: MCP turns Claude from a language model into an agent that can take actions on your system. The model doesn't change. The capability does.

The most useful MCP servers

All official MCP servers are open source at github.com/modelcontextprotocol/servers. There are hundreds of community servers for Slack, Notion, Linear, Jira, and more.

Before you install: Node.js

Most MCP servers are Node.js packages. You need Node 18+ installed. Check if you have it:

node --version
# Should output v18.x.x or higher

# If you don't have Node, install it:
# macOS: brew install node
# Windows: download from nodejs.org

You also need npx, which comes with Node. The MCP servers run via npx — no global install required.

Install the filesystem server

The filesystem MCP server gives Claude read and write access to directories you specify. This is the server you'll use most.

Step 1: Find your Claude Desktop config file

# macOS
~/Library/Application Support/Claude/claude_desktop_config.json

# Windows
%APPDATA%\Claude\claude_desktop_config.json

Open this file in any text editor. If it doesn't exist, create it.

Step 2: Add the filesystem server config

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/yourname/Documents",
        "/Users/yourname/Projects"
      ]
    }
  }
}

Replace the paths with the directories you want Claude to access. You can add multiple directories. The server will only access those directories — nothing else.

Step 3: Restart Claude Desktop

Quit Claude Desktop completely and reopen it. When it starts, it will connect to the filesystem server. You'll see a small hammer icon (🔨) in the conversation input — that means MCP tools are available.

Connect Claude to your GitHub repos

The GitHub MCP server lets Claude read your code, browse issues, create PRs, and more. You'll need a GitHub personal access token.

Create a GitHub token

Go to GitHub → Settings → Developer settings → Personal access tokens → Fine-grained tokens. Create a token with these permissions: Contents (read/write), Issues (read/write), Pull requests (read/write). Copy the token.

Add to your config

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/yourname/Projects"
      ]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

Put MCP to work

With filesystem MCP installed, try these prompts on your own files:

1. "Read the file at /path/to/your/document.txt and summarize it in 3 bullets."

2. "Look at the files in /path/to/your/project and give me a high-level overview of the structure."

3. "Read my README.md and identify any sections that are unclear or missing."

4. "Read the last 3 files I modified in /path/to/your/folder."

5. (If using GitHub MCP)
   "List the open issues in my [repo-name] repository and group them by theme."

The first time Claude reads your actual files and responds with real, specific information about them, the shift in capability becomes obvious. This is what makes the desktop app different.