Filesystem MCP Server

A local filesystem MCP server that lets LLMs and AI agents read, write, search, diff, patch, and manage files safely and efficiently. Built for reliable, structured, and controlled filesystem interaction.

Overview

A secure, production-ready Model Context Protocol server that gives AI assistants controlled access to the local filesystem. All operations are sandboxed to explicitly allowed directories with path traversal prevention, sensitive file blocking, and optional Bearer token authentication.

Supports stdio (default) and Streamable HTTP transport with SSE support. HTTP sessions are implemented with isolated per-session server state.

Key Features

• 18 filesystem tools — read, write, search, diff, patch, hash, and bulk operations with structured output schemas
• Security-first — path validation, symlink escape prevention, sensitive file denylist, localhost-only CORS, optional API key auth
• Dual transport — stdio for local use, Streamable HTTP with SSE for networked/multi-session deployments
• Structured output — all tools return typed outputSchema / structuredContent for reliable LLM parsing
• Self-documenting — 6 built-in resources (internal://instructions, internal://tool-catalog, etc.) and 4 built-in prompts (get-help, compare-files, analyze-path, get-tool-help)

Requirements

• Node.js >= 24

Quick Start

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@j0hanz/filesystem-mcp@latest"]
    }
  }
}

Docker

docker run -i --rm -v /path/to/project:/workspace:ro ghcr.io/j0hanz/filesystem-mcp /workspace

Or using Docker Compose:

services:
  filesystem-mcp:
    build: .
    stdin_open: true
    volumes:
      - ./:/projects/workspace:ro
    command: ['/projects/workspace']

CLI Usage

filesystem-mcp [options] [allowedDirs...]

Arguments:
  allowedDirs              

... [View full README on GitHub](https://github.com/j0hanz/filesystem-mcp#readme)