MCP File System Server
A secure, sandboxed file system server that enables reading, writing, searching, and managing files through MCP-compatible AI clients with path traversal protection and size limits.
README
📁 MCP File System Server
A secure, sandboxed file system server for the Model Context Protocol. Read, write, search, and manage files through any MCP-compatible AI client — with path traversal protection, hidden file controls, and size limits.
✨ Features
- 🔧 5 Tools:
read_file,write_file,list_directory,search_files,get_file_info - 📄 1 Resource:
file://{path}— read files via MCP resource URIs - 🛡️ Security-First Architecture: Sandbox isolation, path traversal prevention, max size limits
- 🔥 Hidden File Control: Block
.hiddenfiles by default, opt-in with--allow-hidden - 🔍 Glob-Based Search: Search by pattern (
*.py,*.md,data_*.csv) - 📊 Rich Metadata: File size, timestamps, permissions, text/binary detection
- 🖥️ Dual Transport: stdio and Streamable HTTP
🚀 Quick Start
# 1. Setup
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
# 2. Create sandbox and run
mkdir -p /tmp/mcp-sandbox
python filesystem_server.py --sandbox /tmp/mcp-sandbox
🛠️ Tools Reference
read_file(path: str) -> str
Read a file and return its contents with metadata (size, modified time, SHA256).
write_file(path: str, content: str) -> str
Write content to a file. Creates parent directories automatically.
list_directory(path: str) -> str
List files and directories with sizes and type indicators.
search_files(pattern: str, base_path: str = "") -> str
Search by glob pattern across the entire sandbox.
get_file_info(path: str) -> str
Get detailed metadata: type, size, timestamps, permissions, text/binary detection.
🛡️ Security Controls
--sandbox PATH # Required: root directory (all operations restricted)
--allow-hidden # Allow access to .hidden files and directories
--max-size N # Max file size in MB (default: 10)
| Attack Vector | Protection |
|---|---|
../../../etc/passwd |
❌ Path traversal prevented |
~/.ssh/id_rsa |
❌ Outside sandbox |
.env files |
❌ Blocked by default |
| 100GB file read | ❌ Size limit (configurable) |
| Hidden directory listing | ❌ Filtered by default |
📄 Resources
file://{path}
Read a file's raw content via MCP resource protocol.
🔌 Connecting to Clients
Claude Desktop
{
"mcpServers": {
"filesystem": {
"command": "python",
"args": ["/ABSOLUTE/PATH/mcp-file-system-server/filesystem_server.py", "--sandbox", "/tmp/mcp-sandbox"]
}
}
}
Cursor
Settings → Features → MCP → Add Server:
- Name:
filesystem - Type:
command - Command:
python /ABSOLUTE/PATH/mcp-file-system-server/filesystem_server.py --sandbox /tmp/mcp-sandbox
📁 Project Structure
mcp-file-system-server/
├── filesystem_server.py # Main server (FastMCP + security)
├── requirements.txt
├── setup.sh
├── README.md
└── .gitignore
🧪 Example Usage
# Read a file
read_file("/tmp/mcp-sandbox/data.txt")
# → "📄 File: /tmp/mcp-sandbox/data.txt\nSize: 1,234 bytes\n..."
# Write a file
write_file("/tmp/mcp-sandbox/output/report.md", "# Report\n\nHello world!")
# → "✅ Written 28 bytes to /tmp/mcp-sandbox/output/report.md"
# Search for Python files
search_files("**/*.py")
# → "🔍 5 file(s) matching **/*.py:\n 📄 scripts/process.py (1,234 bytes)\n..."
📜 License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。