FastMCP File Server

A versatile, secure file server implementing the Model Context Protocol (MCP) that provides AI assistants with safe file operations. Features multiple connection modes, configurable access levels, and comprehensive security controls for various deployment scenarios.

🚀 Features

• Comprehensive File Operations: Create, read, write, delete, copy, move, rename files and directories

• Advanced Text Manipulation: Line-specific operations, search and replace, pattern matching

• File Analysis: Size, permissions, timestamps, hash verification, diff generation

• Batch Operations: Handle multiple files efficiently in single operations

• Archive Support: Create and extract ZIP files

• Format Conversion: Text to PDF, image format conversion, CSV ↔ JSON

• Multiple Connection Modes: stdio, HTTP, and public access via ngrok

• Tiered Access Control: Read-only, Read/Write, and Admin permission levels

• Security First: All operations restricted to configured safe directories

Related MCP server: MCP Filesystem Server

📦 Installation

From PyPI (Recommended)

# Using uv (recommended)
uv tool install fastmcp-file-server

# Using pip
pip install fastmcp-file-server

From Source

git clone https://github.com/Luxshan2000/Local-File-MCP-Server.git
cd Local-File-MCP-Server
uv sync

🔧 Quick Start

Basic Usage

# Set allowed directory
export MCP_ALLOWED_PATH="/path/to/your/files"

# Start stdio server (for Claude Desktop)
fastmcp-file-server

# Start HTTP server
fastmcp-file-server-http

# Start HTTP server bypassing security warning (not recommended)
fastmcp-file-server-http --ignore-keys

With Authentication

# Set admin key for HTTP mode
export MCP_ADMIN_KEY="your-secret-token"
export MCP_HTTP_PORT=8082
fastmcp-file-server-http

⚙️ Configuration

Environment Variables

Configuration Files

Create a .env file in your project root:

# Required: Safe directory for file operations
MCP_ALLOWED_PATH=/absolute/path/to/your/files

# Optional: HTTP server settings
MCP_HTTP_PORT=8082

# Optional: Multi-tier authentication tokens
MCP_READ_KEY=readonly-token-here
MCP_WRITE_KEY=readwrite-token-here  
MCP_ADMIN_KEY=admin-token-here

# Optional: File restrictions
MCP_MAX_FILE_SIZE=10485760
MCP_ALLOWED_EXTENSIONS=.txt,.json,.md,.csv,.log,.xml,.yaml,.yml,.conf,.cfg,.zip,.pdf,.jpg,.png

🔗 Integration

Claude Desktop Integration

Configuration file locations:

• macOS: ~/Library/Application Support/Claude/config.json

• Windows: %APPDATA%\Claude\config.json

Stdio Mode (Direct Integration)

{
  "mcpServers": {
    "local-file-server": {
      "command": "fastmcp-file-server",
      "env": {
        "MCP_ALLOWED_PATH": "/absolute/path/to/your/allowed/directory"
      }
    }
  }
}

HTTP Mode (Local Server)

1. Start the HTTP server:

export MCP_ADMIN_KEY="your-secret-token"
fastmcp-file-server-http

2. Configure Claude Desktop:

{
  "mcpServers": {
    "local-file-server-http": {
      "transport": "http",
      "url": "http://127.0.0.1:8082/mcp",
      "headers": {
        "Authorization": "Bearer your-secret-token"
      }
    }
  }
}

HTTP Mode with mcp-remote Proxy

For environments requiring a proxy:

# Install mcp-remote
npm install -g mcp-remote

{
  "mcpServers": {
    "local-file-server-proxy": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://127.0.0.1:8082/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer your-secret-token"
      }
    }
  }
}

Public Access with ngrok

For web-based AI systems (ChatGPT, etc.):

# Terminal 1: Start authenticated HTTP server
export MCP_ADMIN_KEY="your-secret-token"
export MCP_HTTP_PORT=8082
fastmcp-file-server-http

# Terminal 2: Expose publicly via ngrok
ngrok http 8082

Use the ngrok URL in your web-based AI system:

• URL: https://abc123.ngrok.io/mcp

• Header: Authorization: Bearer your-secret-token

🔒 Security

Security Features

⚠️ HTTP Mode Security Warning:

When starting the HTTP server without any authentication tokens configured, the system will display a security warning and prompt for confirmation. This prevents accidentally running an unprotected server.

# This will trigger a security warning:
fastmcp-file-server-http

# To bypass the warning (not recommended):
fastmcp-file-server-http --ignore-keys

Token Management

⚠️ Important Security Notes:

• With Keys: When ANY token is set (MCP_READ_KEY, MCP_WRITE_KEY, or MCP_ADMIN_KEY), all HTTP requests require the Authorization: Bearer <token> header

• Without Keys: If NO tokens are set, the server runs without authentication (use only in secure environments)

• Multi-tier Access: Different tokens provide different permission levels

• Temporary Exposure: For ngrok or temporary remote access, always use strong tokens and revoke access when done

• Key Rotation: Regularly rotate your tokens, especially after temporary exposures

Access Levels

• No Tokens Set: Server runs without authentication (stdio mode safe, HTTP local only)

• Read Token: MCP_READ_KEY - File listing, reading, searching, comparison operations

• Write Token: MCP_WRITE_KEY - All read operations plus create, modify, copy, move, convert

• Admin Token: MCP_ADMIN_KEY - All operations including file and directory deletion

Best Practices

1. Never commit secrets: Use .env files (added to .gitignore)

2. Use strong tokens: Generate cryptographically secure random tokens (openssl rand -hex 32)

3. Limit access scope: Set MCP_ALLOWED_PATH to the minimum required directory

4. Choose appropriate token level: Use read-only tokens for analysis, admin only when deletion needed

5. Monitor usage: Check logs for unauthorized access attempts

6. Temporary access: Unset all tokens and restart after temporary exposures

💡 Usage Examples

File Operations

# Basic operations
"Create a file called notes.txt with my meeting notes"
"Read lines 10-20 from config.py"
"Copy config.json to backup/config_backup.json"

# Advanced operations
"Search for 'TODO' comments in all Python files"
"Replace 'old_function' with 'new_function' in utils.py"
"Create a ZIP archive of all source files"
"Convert report.txt to PDF format"
"Calculate SHA256 hash of important_file.pdf"

Batch Operations

"Read all .py files in the src/ directory"
"Create these 5 configuration files with their content"
"Delete all .tmp files in the workspace"
"Find all JavaScript files containing 'console.log'"

🛠️ Development

See DEVELOPER.md for detailed development setup and contribution guidelines.

Quick Development Setup

# Clone repository
git clone https://github.com/Luxshan2000/Local-File-MCP-Server.git
cd Local-File-MCP-Server

# Install dependencies
uv sync

# Run development server
uv run server          # stdio mode
uv run server-http     # HTTP mode

# Run tests and linting
uv run test
uv run lint
uv run format

📊 API Reference

Available Tools

🐛 Troubleshooting

Common Issues

Server won't start:

# Reinstall dependencies
uv sync

# Check Python version
python --version  # Requires Python 3.10+

Claude Desktop not connecting:

1. Verify all paths in configuration are absolute (full paths)

2. Restart Claude Desktop after changing configuration

3. Check server starts without errors: uv run server

4. Ensure MCP_ALLOWED_PATH directory exists and is accessible

HTTP authentication fails:

1. Verify MCP_ADMIN_KEY is set before starting server

2. Check Authorization header format: Bearer your-secret-token

3. Ensure token matches exactly (no extra spaces)

Permission denied errors:

1. Check file/directory permissions

2. Verify MCP_ALLOWED_PATH is accessible

3. Ensure user has read/write permissions in the allowed directory

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🤝 Contributing

We welcome contributions! Please see DEVELOPER.md for development setup and CODE_OF_CONDUCT.md for community guidelines.

🔗 Links

• Repository: https://github.com/Luxshan2000/Local-File-MCP-Server

• PyPI Package: https://pypi.org/project/fastmcp-file-server/

• Issues: https://github.com/Luxshan2000/Local-File-MCP-Server/issues

• Model Context Protocol: https://modelcontextprotocol.io/

⭐ Support

If you find this project useful, please consider giving it a star on GitHub!