MCP Filesystem Server

Configure and use the Model Context Protocol (MCP) filesystem server to enable AI assistants to perform secure file and directory operations on your system.

What This Skill Does

This skill helps you set up and use the `@agent-infra/mcp-server-filesystem` package, which provides AI agents with controlled access to filesystem operations through the Model Context Protocol. It supports multiple deployment modes and AI development environments.

Prerequisites

Installation Instructions

For Local (Stdio) Setup

1. **Identify your MCP client's configuration file location:**

- VS Code/Cursor: `.vscode/mcp.json` or global settings

- Windsurf: Follow their MCP documentation

- Claude Desktop: `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or equivalent on other platforms

2. **Add the filesystem server configuration:**

```json

{

"mcpServers": {

"filesystem": {

"command": "npx",

"args": [

"@agent-infra/mcp-server-filesystem@latest"

]

}

}

}

```

3. **Configure allowed directories (recommended for security):**

Add the `--allowed-directories` flag to restrict access to specific paths:

```json

{

"mcpServers": {

"filesystem": {

"command": "npx",

"args": [

"@agent-infra/mcp-server-filesystem@latest",

"--allowed-directories",

"/path/to/project1",

"/path/to/project2"

]

}

}

}

```

4. **Restart your MCP client** to load the new configuration.

For VS Code via CLI

Run this command in your terminal:

```bash

code --add-mcp '{"name":"filesystem","command":"npx","args":["@agent-infra/mcp-server-filesystem@latest"]}'

```

For Remote (SSE/HTTP) Setup

1. **Start the MCP server on a specific port:**

```bash

npx @agent-infra/mcp-server-filesystem --port 8089 --allowed-directories /path/to/allowed/dir

```

2. **Configure your MCP client with the remote endpoint:**

For SSE:

```json

{

"mcpServers": {

"filesystem": {

"url": "http://127.0.0.1:8089/sse"

}

}

}

```

For Streamable HTTP (recommended):

```json

{

"mcpServers": {

"filesystem": {

"type": "streamable-http",

"url": "http://127.0.0.1:8089/mcp"

}

}

}

```

For JavaScript/TypeScript In-Memory Integration

If building a custom MCP client:

```javascript

import { Client } from '@modelcontextprotocol/sdk/client/index.js';

import { InMemoryTransport } from '@modelcontextprotocol/sdk/inMemory.js';

import { createServer } from '@agent-infra/mcp-server-filesystem';

const client = new Client(

{

name: 'filesystem-client',

version: '1.0',

},

{

capabilities: {},

},

);

const server = createServer();

const [clientTransport, serverTransport] = InMemoryTransport.createLinkedPair();

await Promise.all([

client.connect(clientTransport),

server.connect(serverTransport),

]);

// List available tools

const tools = await client.listTools();

// Call a tool

const result = await client.callTool({

name: 'list_directory',

arguments: {

path: '/your/path'

},

});

```

Available Tools

Once configured, the AI assistant will have access to filesystem tools such as:

Security Best Practices

1. **Always use `--allowed-directories`** to restrict filesystem access to specific paths

2. **Avoid granting access to:**

- System directories (`/etc`, `/System`, `/Windows`)

- User home directory root (unless specifically needed)

- Directories containing sensitive credentials

3. **Use separate configurations** for different projects with appropriate directory restrictions

4. **Review the AI's file operations** before confirming destructive actions

Usage Examples

After setup, you can ask your AI assistant:

Troubleshooting

Notes