Local MCP Servers

In the last lesson, you connected Claude to cloud services using Connectors. That is great for tools like Google Drive, Notion, and Slack that live on the internet. But what about the files on your actual computer? Your notes, your code, your local documents?

That is where MCP comes in. MCP stands for Model Context Protocol, and it is the technology that lets Claude talk to tools and data sources on your local machine. If Connectors are like giving Claude phone lines to cloud services, MCP servers are like plugging USB cables directly into your computer.

This lesson is a bit more hands-on than the previous ones. You will be editing a configuration file in a text editor. But do not worry; the process is straightforward and we will walk through every step.

The USB Port Analogy

Here is the simplest way to think about MCP:

Your computer has USB ports. You can plug in a keyboard, a mouse, a camera, an external drive. Each device has its own purpose, but they all use the same standard connector (USB). Your computer does not need to know in advance what you will plug in. It just needs to support the USB protocol.

MCP works the same way for AI:

• Claude Desktop is your computer (the MCP client)
• MCP servers are the USB devices you plug in
• The MCP protocol is the USB standard that makes everything work together

Just like USB, MCP is an open standard. Anyone can build an MCP server for any tool or data source. Anthropic created the protocol, but the community builds the servers. There are already MCP servers for filesystems, databases, APIs, note-taking apps, browsers, and much more.

Finding Your Config File

Before you can add MCP servers, you need to know where Claude Desktop keeps its configuration file. The location depends on your operating system.

macOS

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

To open this in Finder:

1. Open Finder
2. Press Cmd + Shift + G (Go to Folder)
3. Paste: ~/Library/Application Support/Claude/
4. Look for claude_desktop_config.json

Windows

%APPDATA%\Claude\claude_desktop_config.json

To find this:

1. Press Win + R to open Run
2. Type %APPDATA%\Claude and press Enter
3. Look for claude_desktop_config.json

What the Config File Looks Like

The file may be empty or contain just {}. Each entry inside mcpServers is a server Claude Desktop will connect to at startup:

{
  "mcpServers": {
    "server-name": {
      "command": "command-to-run",
      "args": ["arg1", "arg2"]
    }
  }
}

Setting Up the Filesystem MCP Server

The filesystem MCP server is the most common starting point. It gives Claude the ability to read and write files in specific directories on your computer. This is incredibly useful for working with local documents, code, and notes.

Prerequisites

You need Node.js installed on your computer. To check:

1. Open your terminal (Terminal on macOS, Command Prompt or PowerShell on Windows)
2. Type: node --version
3. If you see a version number (like v20.11.0), you are good
4. If you get an error, download Node.js from nodejs.org (choose the LTS version)

You will also need npx, which comes bundled with Node.js. Verify with npx --version.

Step-by-Step Setup

code ~/Library/Application\ Support/Claude/claude_desktop_config.json

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

Setting Up the Obsidian MCP Server

If you use Obsidian for note-taking, there is an MCP server that gives Claude direct access to your vault. This is powerful because Claude can search your notes, read specific pages, and help you write new ones, all without leaving the conversation.

Configuration

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "obsidian": {
      "command": "npx",
      "args": [
        "-y",
        "obsidian-mcp-server",
        "/path/to/your/obsidian/vault"
      ]
    }
  }
}

Replace /path/to/your/obsidian/vault with the actual path to your Obsidian vault folder. If you are unsure where it is, open Obsidian, go to Settings, and look at the vault path.

Multiple Servers at Once

You can run multiple MCP servers simultaneously. Here is what a config file with both filesystem and Obsidian looks like:

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

Each server runs independently. Claude can use tools from all connected servers in the same conversation.

Other Popular MCP Servers

The MCP ecosystem is growing rapidly. Here are a few examples -- notice how they all follow the same config pattern:

{
  "mcpServers": {
    "sqlite": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-sqlite", "/path/to/database.db"]
    },
    "git": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-git", "--repository", "/path/to/repo"]
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-server-brave-search"],
      "env": { "BRAVE_API_KEY": "your-api-key-here" }
    }
  }
}

The sqlite server lets Claude query local databases. The git server gives access to history, diffs, and branches. The brave-search server demonstrates the env field -- some servers need API keys or environment variables passed in their configuration.

Troubleshooting

MCP is still relatively new, and things can go wrong. Here are the most common issues and how to fix them.

"Server not found" or "Failed to start"

Cause: Usually a problem with the command path or Node.js installation.

Fix:

1. Run the command manually in your terminal to see the actual error: npx -y @modelcontextprotocol/server-filesystem /Users/yourname/Documents
2. If npx is not found, it may not be in your PATH. Find the full path with which npx (macOS) or where npx (Windows) and use it in your config:

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

"Permission denied"

Cause: Claude (via the MCP server) does not have permission to access the specified directory.

Fix:

• Make sure the directory exists
• Make sure your user account has read access to the directory
• On macOS, you may need to grant Terminal (or Node.js) Full Disk Access in System Settings > Privacy & Security

Claude does not show MCP tools

Cause: The config file is not being read properly.

Fix:

1. Double-check the file name: it must be exactly claude_desktop_config.json
2. Verify it is valid JSON (no trailing commas, proper quoting). You can paste it into jsonlint.com to validate.
3. Check the file location matches your OS (see the paths above)
4. Make sure you fully quit and restarted Claude Desktop (not just closed the window)

Server crashes after a few minutes

Cause: Often a memory or timeout issue with the MCP server.

Fix:

• Update the MCP server package: npx -y @modelcontextprotocol/server-filesystem@latest
• Check for open issues on the server's GitHub repository
• Try limiting the scope (fewer directories, smaller database)

MCP Security Considerations

MCP servers run on your local machine with your user permissions. This means:

• A filesystem MCP server can read any file you can read
• A database MCP server can query any database you point it at
• A server with write access can modify files

This is powerful but requires responsibility.

Best practices:

• Only install MCP servers from trusted sources
• Limit directory access to specific folders (never your entire disk)
• Review the server's code on GitHub if you are unsure
• Be cautious with MCP servers that request write access
• On shared machines, be aware that your config file contains your server setup

How MCP Fits Into Your Toolkit

| Layer | What | When to Use | |-------|------|-------------| | Custom Instructions | Text in your Project settings | Persistent rules and context | | Knowledge Base | Uploaded documents | Static reference materials | | Connectors | Cloud service integrations | Live data from web services | | MCP Servers | Local tool connections | Files, databases, and tools on your machine |

Together, these four layers give Claude comprehensive access to your world. You do not need all four for every task, but knowing when to reach for each one is what separates a casual Claude user from someone who has built a truly effective AI workflow.