Filesystem MCP Server (@sylphlab/filesystem-mcp)

npm version Docker Pulls

Empower your AI agents (like Cline/Claude) with secure, efficient, and token-saving access to your project files. This Node.js server implements the Model Context Protocol (MCP) to provide a robust set of filesystem tools, operating safely within a defined project root directory.

Installation

There are several ways to use the Filesystem MCP Server:

1. Recommended:

The simplest way is via npx or bunx, configured directly in your MCP host environment (e.g., Roo/Cline's mcp_settings.json). This ensures you always use the latest version from npm without needing local installation or Docker.

Example (

{ "mcpServers": { "filesystem-mcp": { "command": "npx", "args": ["@sylphlab/filesystem-mcp"], "name": "Filesystem (npx)" } } }

Example (

{ "mcpServers": { "filesystem-mcp": { "command": "bunx", "args": ["@sylphlab/filesystem-mcp"], "name": "Filesystem (bunx)" } } }

Important: The server uses its own Current Working Directory (cwd) as the project root. Ensure your MCP Host (e.g., Cline/VSCode) is configured to launch the command with the cwd set to your active project's root directory.

2. Docker

Use the official Docker image for containerized environments.

Example MCP Host Configuration:

{ "mcpServers": { "filesystem-mcp": { "command": "docker", "args": [ "run", "-i", "--rm", "-v", "/path/to/your/project:/app", // Mount your project to /app "sylphlab/filesystem-mcp:latest" ], "name": "Filesystem (Docker)" } } }

Remember to replace

3. Local Build (For Development)

Clone: git clone https://github.com/sylphlab/filesystem-mcp.git
Install: cd filesystem-mcp && pnpm install (Using pnpm now)
Build: pnpm run build
Configure MCP Host:
{ "mcpServers": { "filesystem-mcp": { "command": "node", "args": ["/path/to/cloned/repo/filesystem-mcp/dist/index.js"], // Updated build dir "name": "Filesystem (Local Build)" } } }
Note: Launch the node command from the directory you intend as the project root.

Related MCP server: drupal-modules-mcp MCP Server

Quick Start

Once the server is configured in your MCP host (see Installation), your AI agent can immediately start using the filesystem tools.

Example Agent Interaction (Conceptual):

Agent: <use_mcp_tool> <server_name>filesystem-mcp</server_name> <tool_name>read_content</tool_name> <arguments>{"paths": ["src/index.ts"]}</arguments> </use_mcp_tool> Server Response: (Content of src/index.ts)

Why Choose This Project?

🛡️ Secure & Convenient Project Root Focus: Operations confined to the project root (cwd at launch).
⚡ Optimized & Consolidated Tools: Batch operations reduce AI-server round trips, saving tokens and latency. Reliable results for each item in a batch.
🚀 Easy Integration: Quick setup via npx/bunx.
🐳 Containerized Option: Available as a Docker image.
🔧 Comprehensive Functionality: Covers a wide range of filesystem tasks.
✅ Robust Validation: Uses Zod schemas for argument validation.

Performance Advantages

(Placeholder: Add benchmark results and comparisons here, demonstrating advantages over alternative methods like individual shell commands.)

Batch Operations: Significantly reduces overhead compared to single operations.
Direct API Usage: More efficient than spawning shell processes for each command.
(Add specific benchmark data when available)

Features

This server equips your AI agent with a powerful and efficient filesystem toolkit:

📁 Explore & Inspect ( List files/directories (recursive, stats), get detailed status for multiple items.
📄 Read & Write Content ( Read/write/append multiple files, creates parent directories.
✏️ Precision Editing & Searching ( Surgical edits (insert, replace, delete) across multiple files with indentation preservation and diff output; regex search with context; multi-file search/replace.
🏗️ Manage Directories ( Create multiple directories including intermediate parents.
🗑️ Delete Safely ( Remove multiple files/directories recursively.
↔️ Move & Copy ( Move/rename/copy multiple files/directories.
🔒 Control Permissions ( Change POSIX permissions and ownership for multiple items.

Key Benefit: All tools accepting multiple paths/operations process each item individually and return a detailed status report.

Design Philosophy

(Placeholder: Explain the core design principles.)

Security First: Prioritize preventing access outside the project root.
Efficiency: Minimize communication overhead and token usage for AI interactions.
Robustness: Provide detailed results and error reporting for batch operations.
Simplicity: Offer a clear and consistent API via MCP.
Standard Compliance: Adhere strictly to the Model Context Protocol.

Comparison with Other Solutions

(Placeholder: Objectively compare with alternatives.)

Feature/Aspect	Filesystem MCP Server	Individual Shell Commands (via Agent)	Other Custom Scripts
Security	High (Root Confined)	Low (Agent needs shell access)	Variable
Efficiency (Tokens)	High (Batching)	Low (One command per op)	Variable
Latency	Low (Direct API)	High (Shell spawn overhead)	Variable
Batch Operations	Yes (Most tools)	No	Maybe
Error Reporting	Detailed (Per item)	Basic (stdout/stderr parsing)	Variable
Setup	Easy (npx/Docker)	Requires secure shell setup	Custom

Future Plans

(Placeholder: List upcoming features or improvements.)

Explore file watching capabilities.
Investigate streaming support for very large files.
Enhance performance for specific operations.
Add more advanced filtering options for list_files.

Documentation

(Placeholder: Add link to the full documentation website once available.)

Full documentation, including detailed API references and examples, will be available at: [Link to Docs Site]

Contributing

Contributions are welcome! Please open an issue or submit a pull request on the GitHub repository.

License

This project is released under the MIT License.

Development

Clone: git clone https://github.com/sylphlab/filesystem-mcp.git
Install: cd filesystem-mcp && pnpm install
Build: pnpm run build (compiles TypeScript to dist/)
Watch: pnpm run dev (optional, recompiles on save)

Publishing (via GitHub Actions)

This repository uses GitHub Actions (.github/workflows/publish.yml) to automatically publish the package to npm and build/push a Docker image to Docker Hub on pushes of version tags (v*.*.*) to the main branch. Requires NPM_TOKEN, DOCKERHUB_USERNAME, and DOCKERHUB_TOKEN secrets configured in the GitHub repository settings.