--- title: How to Install the Filesystem MCP Server description: The filesystem MCP server gives your AI apps read and write access to any folder on your machine. Here is how to install it in under 5 minutes using NPX. author: FractionalSkill ---

How to Install the Filesystem MCP Server

Most MCP servers connect your AI tools to external services. The filesystem server does something different. It connects them to your own machine. Folders, files, notes, documents -- anything stored locally becomes readable and writable directly from Claude Desktop or any other MCP-compatible app.

For operators who keep client deliverables, notes, and project files on their hard drive, this is one of the most immediately useful servers available.

What the filesystem server actually does

The filesystem server is a Node.js server that implements MCP for local file operations. Here is what it can handle:

• Read files -- Pull the contents of any file in your allowed folders
• Write and create files -- Generate new files or edit existing ones
• Move files -- Reorganize folders without leaving your AI session
• Search files -- Find files by name or content across your allowed directories
• List directories -- Get a full picture of what's in a folder

The key phrase is "allowed folders." You specify exactly which directories the server can access when you configure it. Nothing outside that list is reachable. That is how access control works with this server.

A practical example: if you manage your notes in a local app like Obsidian, every file in that vault is a plain markdown file sitting on your hard drive. Wire the filesystem server to that vault and your AI tools can query notes, create new ones, and list files across directories -- all without touching a cloud sync.

What you need before installing

The filesystem server uses NPX to run, which means you need Node.js and NPM installed on your machine. If you already completed the MCP development environment setup, you have both.

If you are not sure, open your terminal and run:

node --version
npm --version

If both return version numbers, you are ready. If not, go to nodejs.org, download the installer for your platform, and run it. NPM comes bundled with Node.js.

> No cloning required. Unlike building a custom server, the filesystem server runs directly via NPX. The command pulls what it needs at runtime. You do not need to download or install a repository anywhere on your machine.

Adding the server to your config

Open Claude Desktop, navigate to Settings, then Developer Settings. This is where your claudedesktopconfig.json lives. Open that file.

Add the filesystem server block inside the mcpServers object. If you have existing servers configured, add this alongside them -- do not overwrite what is already there.

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

The folder paths at the end of the args array are your allow list. Replace those paths with the actual directories you want the server to access. Add as many as you need, separated by commas. Each one becomes accessible to Claude.

Copy your exact path. In Finder on Mac, right-click any folder and select "Copy as Pathname" to get the full path without typos. On Windows, hold Shift, right-click the folder, and select "Copy as path."

Save the file and restart Claude Desktop. The server initializes on startup.

Verifying it works

After restarting, open a new conversation in Claude Desktop and look at the available tools. You should see tools coming from the filesystem server -- things like readfile, writefile, listdirectory, and searchfiles. You can click through to read each tool's description.

To confirm everything is connected, ask Claude what files are in one of your allowed folders:

What files are in /Users/yourname/Documents/client-work?

Claude will call the list_directory tool and return the contents. If you see the file list, the server is working.

From there, try something more substantive. Ask Claude to read a specific file and summarize it. Ask it to check what dependencies are listed in a package.json. Ask it to create a new file. Each of these uses a different tool from the filesystem server, and Claude chooses which one to call based on what you ask.

> Permission prompts are expected. Claude Desktop will ask your approval before each tool call by default. That is normal behavior. If you approve a tool type frequently, you can adjust permission settings in Claude Desktop. Start with approvals on so you can see exactly what the server is doing.

What this changes for your workflow

The filesystem server shifts what your AI tools can actually reach. Instead of copying and pasting file contents into a chat window, Claude can read the source directly. Instead of manually creating files based on AI output, Claude can write them for you.

For operators managing client work across multiple local folders, this translates to faster document processing, automated file creation as part of deliverable workflows, and AI access to the notes and context you already maintain. The allow list keeps it scoped to exactly what you decide to expose.

> Expand incrementally. Start with one folder you use actively, get familiar with how the tools behave, then add more directories as you find use cases. This keeps your access control intentional rather than broad.