Skip to main content
Glama

Filesystem MCP Server

TypeScript Model Context Protocol Version License Status GitHub

Empower your AI agents with robust, platform-agnostic file system capabilities, now with STDIO & Streamable HTTP transport options.

This Model Context Protocol (MCP) server provides a secure and reliable interface for AI agents to interact with the local filesystem. It enables reading, writing, updating, and managing files and directories, backed by a production-ready TypeScript foundation featuring comprehensive logging, error handling, security measures, and now supporting both STDIO and HTTP transports.

Table of Contents

Related MCP server: MCP Filesystem Server

Overview

The Model Context Protocol (MCP) is a standard framework allowing AI models to securely interact with external tools and data sources (resources). This server implements the MCP standard to expose essential filesystem operations as tools, enabling AI agents to:

  • Read and analyze file contents.

  • Create, modify, or overwrite files.

  • Manage directories and file paths.

  • Perform targeted updates within files.

Built with TypeScript, the server emphasizes type safety, modularity, and robust error handling, making it suitable for reliable integration into AI workflows. It now supports both STDIO for direct process communication and HTTP for network-based interactions.

Architecture

The server employs a layered architecture for clarity and maintainability:

flowchart TB
    subgraph TransportLayer["Transport Layer"]
        direction LR
        STDIO["STDIO Transport"]
        HTTP["HTTP Transport (Express, JWT Auth)"]
    end

    subgraph APILayer["API Layer"]
        direction LR
        MCP["MCP Protocol Interface"]
        Val["Input Validation (Zod)"]
        PathSan["Path Sanitization"]

        MCP --> Val --> PathSan
    end

    subgraph CoreServices["Core Services"]
        direction LR
        Config["Configuration (Zod-validated Env Vars)"]
        Logger["Logging (Winston, Context-aware)"]
        ErrorH["Error Handling (McpError, ErrorHandler)"]
        ServerLogic["MCP Server Logic"]
        State["Session State (Default Path)"]

        Config --> ServerLogic
        Logger --> ServerLogic & ErrorH
        ErrorH --> ServerLogic
        State --> ServerLogic
    end

    subgraph ToolImpl["Tool Implementation"]
        direction LR
        FSTools["Filesystem Tools"]
        Utils["Core Utilities (Internal, Security, Metrics, Parsing)"]

        FSTools --> ServerLogic
        Utils -- Used by --> FSTools
        Utils -- Used by --> CoreServices
        Utils -- Used by --> APILayer
    end

    TransportLayer --> MCP
    PathSan --> FSTools

    classDef layer fill:#2d3748,stroke:#4299e1,stroke-width:3px,rx:5,color:#fff
    classDef component fill:#1a202c,stroke:#a0aec0,stroke-width:2px,rx:3,color:#fff
    class TransportLayer,APILayer,CoreServices,ToolImpl layer
    class STDIO,HTTP,MCP,Val,PathSan,Config,Logger,ErrorH,ServerLogic,State,FSTools,Utils component
  • Transport Layer: Handles communication via STDIO or HTTP (with Express.js and JWT authentication).

  • API Layer: Manages MCP communication, validates inputs using Zod, and sanitizes paths.

  • Core Services: Oversees configuration (Zod-validated environment variables), context-aware logging, standardized error reporting, session state (like the default working directory), and the main MCP server instance.

  • Tool Implementation: Contains the specific logic for each filesystem tool, leveraging a refactored set of shared utilities categorized into internal, security, metrics, and parsing modules.

Features

  • Comprehensive File Operations: Tools for reading, writing, listing, deleting, moving, and copying files and directories.

  • Targeted Updates: update_file tool allows precise search-and-replace operations within files, supporting plain text and regex.

  • Session-Aware Path Management: set_filesystem_default tool establishes a default working directory for resolving relative paths during a session.

  • Dual Transport Support:

    • STDIO: For direct, efficient communication when run as a child process.

    • HTTP: For network-based interaction, featuring RESTful endpoints, Server-Sent Events (SSE) for streaming, and JWT-based authentication.

  • Security First:

    • Built-in path sanitization prevents directory traversal attacks.

    • JWT authentication for HTTP transport.

    • Input validation with Zod.

  • Robust Foundation: Includes production-grade utilities, now reorganized for better modularity:

    • Internal Utilities: Context-aware logging (Winston), standardized error handling (McpError, ErrorHandler), request context management.

    • Security Utilities: Input sanitization, rate limiting, UUID and prefixed ID generation.

    • Metrics Utilities: Token counting.

    • Parsing Utilities: Natural language date parsing, partial JSON parsing.

  • Enhanced Configuration: Zod-validated environment variables for type-safe and reliable setup.

  • Type Safety: Fully implemented in TypeScript for improved reliability and maintainability.

Installation

Steps

  1. Clone the repository:

    git clone https://github.com/cyanheads/filesystem-mcp-server.git
    cd filesystem-mcp-server
  2. Install dependencies:

    npm install
  3. Build the project:

    npm run build

    This compiles the TypeScript code to JavaScript in the dist/ directory and makes the main script executable. The executable will be located at dist/index.js.

Configuration

Configure the server using environment variables (a .env file is supported):

Core Server Settings:

  • MCP_LOG_LEVEL (Optional): Minimum logging level (e.g., debug, info, warn, error). Defaults to debug.

  • LOGS_DIR (Optional): Directory for log files. Defaults to ./logs in the project root.

  • NODE_ENV (Optional): Runtime environment (e.g., development, production). Defaults to development.

Transport Settings:

  • MCP_TRANSPORT_TYPE (Optional): Communication transport (stdio or http). Defaults to stdio.

    • If http is selected:

      • MCP_HTTP_PORT (Optional): Port for the HTTP server. Defaults to 3010.

      • MCP_HTTP_HOST (Optional): Host for the HTTP server. Defaults to 127.0.0.1.

      • MCP_ALLOWED_ORIGINS (Optional): Comma-separated list of allowed CORS origins (e.g., http://localhost:3000,https://example.com).

      • MCP_AUTH_SECRET_KEY (Required for HTTP Auth): A secure secret key (at least 32 characters long) for JWT authentication. CRITICAL for production.

Filesystem Security:

  • FS_BASE_DIRECTORY (Optional): Defines the root directory for all filesystem operations. This can be an absolute path or a path relative to the project root (e.g., ./data_sandbox). If set, the server's tools will be restricted to accessing files and directories only within this specified (and resolved absolute) path and its subdirectories. This is a crucial security feature to prevent unintended access to other parts of the filesystem. If not set (which is not recommended for production environments), a warning will be logged, and operations will not be restricted.

LLM & API Integration (Optional):

  • OPENROUTER_APP_URL: Your application's URL for OpenRouter.

  • OPENROUTER_APP_NAME: Your application's name for OpenRouter. Defaults to MCP_SERVER_NAME.

  • OPENROUTER_API_KEY: API key for OpenRouter services.

  • LLM_DEFAULT_MODEL: Default LLM model to use (e.g., google/gemini-2.5-flash-preview-05-20).

  • LLM_DEFAULT_TEMPERATURE, LLM_DEFAULT_TOP_P, LLM_DEFAULT_MAX_TOKENS, LLM_DEFAULT_TOP_K, LLM_DEFAULT_MIN_P: Default parameters for LLM calls.

  • GEMINI_API_KEY: API key for Google Gemini services.

OAuth Proxy Integration (Optional, for advanced scenarios):

  • OAUTH_PROXY_AUTHORIZATION_URL, OAUTH_PROXY_TOKEN_URL, OAUTH_PROXY_REVOCATION_URL, OAUTH_PROXY_ISSUER_URL, OAUTH_PROXY_SERVICE_DOCUMENTATION_URL, OAUTH_PROXY_DEFAULT_CLIENT_REDIRECT_URIS: Configuration for an OAuth proxy.

Refer to src/config/index.ts and the .clinerules file for the complete list and Zod schema definitions.

Usage with MCP Clients

To allow an MCP client (like an AI assistant) to use this server:

  1. Run the Server: Start the server from your terminal:

    node dist/index.js
    # Or if you are in the project root:
    # npm start
  2. Configure the Client: Add the server to your MCP client's configuration. The exact method depends on the client.

    For STDIO Transport (Default): Typically involves specifying:

    • Command: node

    • Arguments: The absolute path to the built server executable (e.g., /path/to/filesystem-mcp-server/dist/index.js).

    • Environment Variables (Optional): Set any required environment variables from the Configuration section.

    Example MCP Settings for STDIO (Conceptual):

    {
      "mcpServers": {
        "filesystem_stdio": {
          "command": "node",
          "args": ["/path/to/filesystem-mcp-server/dist/index.js"],
          "env": {
            "MCP_LOG_LEVEL": "debug"
            // Other relevant env vars
          },
          "disabled": false,
          "autoApprove": []
        }
      }
    }

    For HTTP Transport: The client will need to know the server's URL (e.g., http://localhost:3010) and how to authenticate (e.g., providing a JWT Bearer token if MCP_AUTH_SECRET_KEY is set). Refer to your MCP client's documentation for HTTP server configuration.

Once configured and running, the client will detect the server and its available tools.

Available Tools

The server exposes the following tools for filesystem interaction:

Tool

Description

set_filesystem_default

Sets a default absolute path for the current session. Relative paths used in subsequent tool calls will be resolved against this default. Resets on server restart.

read_file

Reads the entire content of a specified file as UTF-8 text. Accepts relative (resolved against default) or absolute paths.

write_file

Writes content to a specified file. Creates the file (and necessary parent directories) if it doesn't exist, or overwrites it if it does. Accepts relative or absolute paths.

update_file

Performs targeted search-and-replace operations within an existing file using an array of {search, replace} blocks. Ideal for localized changes. Supports plain text or regex search (useRegex: true) and replacing all occurrences (replaceAll: true). Accepts relative or absolute paths. File must exist.

list_files

Lists files and directories within a specified path. Options include recursive listing (includeNested: true) and limiting the number of entries (maxEntries). Returns a formatted tree structure. Accepts relative or absolute paths.

delete_file

Permanently removes a specific file. Accepts relative or absolute paths.

delete_directory

Permanently removes a directory. Use recursive: true to remove non-empty directories and their contents (use with caution!). Accepts relative or absolute paths.

create_directory

Creates a new directory at the specified path. By default (create_parents: true), it also creates any necessary parent directories. Accepts relative or absolute paths.

move_path

Moves or renames a file or directory from a source path to a destination path. Accepts relative or absolute paths for both.

copy_path

Copies a file or directory from a source path to a destination path. For directories, it copies recursively by default (recursive: true). Accepts relative or absolute paths.

Refer to the tool registration files (src/mcp-server/tools/*/registration.ts) for detailed input/output schemas (Zod/JSON Schema).

Project Structure

The codebase is organized for clarity and maintainability:

filesystem-mcp-server/
├── dist/                 # Compiled JavaScript output (after npm run build)
├── logs/                 # Log files (created at runtime)
├── node_modules/         # Project dependencies
├── src/                  # TypeScript source code
│   ├── config/           # Configuration loading (index.ts)
│   ├── mcp-server/       # Core MCP server logic
│   │   ├── server.ts     # Server initialization, tool registration, transport handling
│   │   ├── state.ts      # Session state management (e.g., default path)
│   │   ├── tools/        # Individual tool implementations (one subdir per tool)
│   │   │   ├── readFile/
│   │   │   │   ├── index.ts
│   │   │   │   ├── readFileLogic.ts
│   │   │   │   └── registration.ts
│   │   │   └── ...       # Other tools (writeFile, updateFile, etc.)
│   │   └── transports/   # Communication transport implementations
│   │       ├── authentication/ # Auth middleware for HTTP
│   │       │   └── authMiddleware.ts
│   │       ├── httpTransport.ts
│   │       └── stdioTransport.ts
│   ├── types-global/     # Shared TypeScript types and interfaces
│   │   ├── errors.ts     # Custom error classes and codes (McpError, BaseErrorCode)
│   │   ├── mcp.ts        # MCP related types
│   │   └── tool.ts       # Tool definition types
│   ├── utils/            # Reusable utility modules, categorized
│   │   ├── internal/     # Core internal utilities (errorHandler, logger, requestContext)
│   │   ├── metrics/      # Metrics-related utilities (tokenCounter)
│   │   ├── parsing/      # Parsing utilities (dateParser, jsonParser)
│   │   ├── security/     # Security-related utilities (idGenerator, rateLimiter, sanitization)
│   │   └── index.ts      # Barrel export for all utilities
│   └── index.ts          # Main application entry point
├── .clinerules           # Cheatsheet for LLM assistants
├── .dockerignore
├── Dockerfile
├── LICENSE
├── mcp.json              # MCP server manifest (generated by SDK or manually)
├── package.json
├── package-lock.json
├── README.md             # This file
├── repomix.config.json
├── smithery.yaml         # Smithery configuration (if used)
└── tsconfig.json         # TypeScript compiler options

For a live, detailed view of the current structure, run: npm run tree (This script might need to be updated if src/scripts/tree.ts was part of the changes).

Developer Note: This repository includes a .clinerules file. This cheat sheet provides your LLM coding assistant with essential context about codebase patterns, file locations, and usage examples. Keep it updated as the server evolves!

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.


Available Tools

10 tools
copy_pathB

Copies a file or directory to a new location. Accepts relative or absolute paths. Defaults to recursive copy for directories.

Inputs: destination_path (required), source_path (required), recursive.

create_directoryB

Creates a directory. Optionally creates parent directories. Accepts relative or absolute paths.

Inputs: path (required), create_parents.

delete_directoryB

Removes a directory. Optionally removes recursively. Accepts relative or absolute paths.

Inputs: path (required), recursive.

delete_fileA

Removes a specific file. Accepts relative or absolute paths.

Inputs: path (required).

list_filesA

Lists files and directories within the specified directory. Optionally lists recursively and returns a tree-like structure. Includes an optional maxEntries parameter (default 50) to limit the number of items returned.

Inputs: path (required), includeNested, maxEntries.

move_pathB

Moves or renames a file or directory. Accepts relative or absolute paths for source and destination.

Inputs: destination_path (required), source_path (required).

read_fileA

Reads the entire content of a specified file as UTF-8 text. Accepts relative or absolute paths. Relative paths are resolved against the session default set by set_filesystem_default.

Inputs: path (required).

set_filesystem_defaultA

Sets a default absolute path for the current session. Relative paths used in other filesystem tools (like readFile) will be resolved against this default. The default is cleared on server restart.

Inputs: path (required).

update_fileA

Performs targeted search-and-replace operations within an existing file using an array of {search, replace} blocks. Preferred for smaller, localized changes. For large-scale updates or overwrites, consider using write_file. Accepts relative or absolute paths. File must exist. Supports optional useRegex (boolean, default false) and replaceAll (boolean, default false).

Inputs: blocks (required), path (required), replaceAll, useRegex.

write_fileA

Writes content to a specified file. Creates the file (and necessary directories) if it doesn't exist, or overwrites it if it does. Accepts relative or absolute paths (resolved like readFile).

Inputs: content (required), path (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 10 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose with no ambiguity: copy_path, move_path, delete_file, and delete_directory handle different file operations; list_files, read_file, write_file, and update_file serve distinct read/write functions; create_directory and set_filesystem_default are unique utilities. The descriptions reinforce these distinctions, making misselection unlikely.

    Naming Consistency4/5

    The tool names follow a consistent snake_case pattern with clear verb_noun structures (e.g., copy_path, delete_file). However, there is a minor deviation with set_filesystem_default, which uses a longer, descriptive name that breaks the simple verb_noun pattern, though it remains readable and consistent in style.

    Tool Count5/5

    With 10 tools, the count is well-scoped for a filesystem server, covering essential operations like create, read, update, delete, list, copy, move, and session management. Each tool earns its place without redundancy, fitting the typical range of 3-15 tools for such a domain.

    Completeness5/5

    The tool set provides complete CRUD/lifecycle coverage for filesystem operations: create_directory, read_file, write_file, update_file, delete_file, and delete_directory handle core file management; list_files, copy_path, and move_path support navigation and organization; set_filesystem_default adds session context. There are no obvious gaps, enabling agents to perform comprehensive filesystem tasks without dead ends.

    Maintenance

    ActivityInactive
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      A Model Context Protocol (MCP) server that allows AI models to safely access and interact with local file systems, enabling reading file contents, listing directories, and retrieving file metadata.
      10 npm
      10
      MIT
    • A
      license
      A
      quality
      F
      maintenance
      A Model Context Protocol server that provides secure and intelligent interaction with files and filesystems, offering smart context management and token-efficient operations for working with large files and complex directory structures.
      21
      67
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      A Model Context Protocol server that extends AI capabilities by providing file system access and management functionalities to Claude or other AI assistants.
      277 npm
      5
      -
    • A
      license
      A
      quality
      D
      maintenance
      A secure Model Context Protocol server that provides controlled filesystem access within predefined directories, enabling AI models to perform file and directory operations with strict path validation.
      16
      81 npm
      7
      MIT