Filesystem MCP Server

Bun-based server implementing Model Context Protocol (MCP) for filesystem operations with comprehensive permission controls and enhanced functionality.

Development uses Bun and the server can run directly from TypeScript with bun, but most MCP clients execute Node-compatible JavaScript. Use node dist/index.js in configs unless you're intentionally running the TypeScript entry with Bun.

Features

Granular permission controls (read-only, full access, or specific operation permissions)
Secure file operations within allowed directories
File operations:
- Read/write/modify files
- Create/list/delete directories
- Move files/directories
- Search files by name or extension
- Get file metadata
Directory operations:
- Tree view of directory structures
- Recursive operations with exclusion patterns
Utility functions:
- XML to JSON conversion
- Multiple file operations in one call
- Advanced file editing with pattern matching
Security features:
- Symlink control
- Path validation
- Sandboxed operations

Note: The server will only allow operations within directories specified via args and according to the configured permissions.

Installation

Install Bun (requires Bun v1.0 or later)

curl -fsSL https://bun.sh/install | bash

Install dependencies
```
bun install
```
Build the project (required for Node runtimes)
```
bun run build
```
Run tests
```
bun test
```

Configuration options

Paths may include environment variables like $HOME, ${CUSTOM}, or %USERPROFILE%. Choose the modality that fits your setup:

Local (Node or Bun)

Use Node for built JavaScript or Bun to run TypeScript directly.

{ "command": "node", "args": ["/path/to/mcp-filesystem/dist/index.js", "$HOME/allowed-directory"] }

{ "command": "bun",  "args": ["/path/to/mcp-filesystem/index.ts", "$HOME/allowed-directory"] }

Git hosted

Run straight from the public repo without cloning.

{ "command": "bunx", "args": ["github:rawr-ai/mcp-filesystem", "$HOME/allowed-directory"] }

{ "command": "npx",  "args": ["github:rawr-ai/mcp-filesystem", "$HOME/allowed-directory"] }

NPM package (coming soon)

Planned publication to rawr-ai/mcp-filesystem.

{ "command": "bunx", "args": ["rawr-ai/mcp-filesystem", "$HOME/allowed-directory"] }

{ "command": "npx",  "args": ["rawr-ai/mcp-filesystem", "$HOME/allowed-directory"] }

Docker

Isolated container environment.

{ "command": "docker", "args": ["run", "--rm", "-v", "$HOME/allowed-directory:/data", "mcp/filesystem", "/data"] }

Hosted service

For managed MCP hosts like glama.ai.

{ "mcpServers": { "filesystem": { "url": "https://glama.ai/rawr-ai/mcp-filesystem" } } }

See the examples/ directory for platform-specific configs (Cursor, Roo, etc.) and additional path variants.

API

Resources

file://system: File system operations interface

Tools

All tool argument schemas are defined with TypeBox and registered via the toolSchemas map in src/schemas. This ensures every tool shares a consistent schema that handlers can reference.

read_file
- Read contents of a file (response-capped)
- Inputs:
  - path (string)
  - maxBytes (number): Maximum bytes to return
- Returns at most maxBytes bytes to protect downstream consumers
read_multiple_files
- Read multiple files simultaneously
- Inputs:
  - paths (string[])
  - maxBytesPerFile (number): Maximum bytes to return per file
- Failed reads won't stop the entire operation
create_file
- Create a new file with content
- Inputs:
  - path (string): File location
  - content (string): File content
- Fails if file already exists
- Requires create permission
modify_file
- Modify an existing file with new content
- Inputs:
  - path (string): File location
  - content (string): New file content
- Fails if file doesn't exist
- Requires edit permission
edit_file
- Make selective edits using pattern matching and formatting
- Features:
  - Line-based and multi-line content matching
  - Whitespace normalization with indentation preservation
  - Multiple simultaneous edits with correct positioning
  - Indentation style detection and preservation
  - Git-style diff output with context
  - Preview changes with dry run mode
- Inputs:
  - path (string): File to edit
  - edits (array): List of edit operations
    - oldText (string): Text to search for (exact match)
    - newText (string): Text to replace with
  - dryRun (boolean): Preview changes without applying (default: false)
  - maxBytes (number): Maximum bytes to read before editing
- Returns detailed diff for dry runs, otherwise applies changes
- Requires edit permission
- Best Practice: Always use dryRun first to preview changes
create_directory
- Create new directory or ensure it exists
- Input: path (string)
- Creates parent directories if needed
- Succeeds silently if directory exists
- Requires create permission
list_directory
- List directory contents with [FILE] or [DIR] prefixes
- Input: path (string)
- Returns detailed listing of files and directories
directory_tree
- Get recursive tree view of directory structure
- Input: path (string)
- Returns JSON structure with files and directories
- Each entry includes name, type, and children (for directories)
move_file
- Move or rename files and directories
- Inputs:
  - source (string): Source path
  - destination (string): Destination path
- Fails if destination exists
- Works for both files and directories
- Requires move permission
delete_file
- Delete a file
- Input: path (string)
- Fails if file doesn't exist
- Requires delete permission
delete_directory
- Delete a directory
- Inputs:
  - path (string): Directory to delete
  - recursive (boolean): Whether to delete contents (default: false)
- Fails if directory is not empty and recursive is false
- Requires delete permission
search_files
- Recursively search for files/directories
- Inputs:
  - path (string): Starting directory
  - pattern (string): Search pattern
  - excludePatterns (string[]): Exclude patterns (glob format supported)
- Case-insensitive matching
- Returns full paths to matches
find_files_by_extension
- Find all files with specific extension
- Inputs:
  - path (string): Starting directory
  - extension (string): File extension to find
  - excludePatterns (string[]): Optional exclude patterns
- Case-insensitive extension matching
- Returns full paths to matching files
get_file_info
- Get detailed file/directory metadata
- Input: path (string)
- Returns:
  - Size
  - Creation time
  - Modified time
  - Access time
  - Type (file/directory)
  - Permissions
get_permissions
- Get current server permissions
- No input required
- Returns:
  - Permission flags (readonly, fullAccess, create, edit, move, delete)
  - Symlink following status
  - Number of allowed directories
list_allowed_directories
- List all directories the server is allowed to access
- No input required
- Returns array of allowed directory paths
xml_to_json
- Convert XML file to JSON format
- Inputs:
  - xmlPath (string): Source XML file
  - jsonPath (string): Destination JSON file
  - maxResponseBytes (number, optional): Maximum size of written JSON; large outputs are summarized
  - options (object, optional):
    - ignoreAttributes (boolean): Skip XML attributes (default: false)
    - preserveOrder (boolean): Maintain property order (default: true)
    - format (boolean): Pretty print JSON (default: true)
    - indentSize (number): JSON indentation (default: 2)
- Requires read permission for XML file
- Requires create or edit permission for JSON file
xml_to_json_string
- Convert XML file to JSON string
- Inputs:
  - xmlPath (string): Source XML file
  - maxResponseBytes (number, optional): Maximum size of returned JSON string; large outputs are summarized
  - options (object, optional):
    - ignoreAttributes (boolean): Skip XML attributes (default: false)
    - preserveOrder (boolean): Maintain property order (default: true)
- Requires read permission for XML file
- Returns JSON string representation (response-capped)
xml_query
- Query XML file using XPath expressions
- Inputs:
  - path (string): Path to the XML file
  - query (string, optional): XPath query to execute
  - structureOnly (boolean, optional): Return only tag structure
  - includeAttributes (boolean, optional): Include attribute info (default: true)
  - maxResponseBytes (number, optional): Maximum size of returned JSON; defaults to 200KB
    - Legacy maxBytes is still accepted and treated as response cap
- XPath examples:
  - Get all elements: //tagname
  - Get elements with specific attribute: //tagname[@attr="value"]
  - Get text content: //tagname/text()
- Parses full file; response is truncated to fit limits as needed
xml_structure
- Analyze XML structure
- Inputs:
  - path (string): Path to the XML file
  - maxDepth (number, optional): How deep to analyze (default: 2)
  - includeAttributes (boolean, optional): Include attribute analysis (default: true)
  - maxResponseBytes (number, optional): Maximum size of returned JSON; defaults to 200KB
    - Legacy maxBytes is still accepted and treated as response cap
- Returns statistical information about elements, attributes, namespaces, and hierarchy
- Parses full file; returns a summarized structure if response exceeds limit
regex_search_content
- Search file contents with a regular expression
- Inputs:
  - path (string): Root directory to search
  - regex (string): Regular expression pattern
  - filePattern (string, optional): Glob to limit files (default: *)
  - maxDepth (number, optional): Directory depth (default: 2)
  - maxFileSize (number, optional): Maximum file size in bytes (default: 10MB)
  - maxResults (number, optional): Maximum number of files with matches (default: 50)
- Returns a human-readable summary of files and matching lines

Argument Validation

The server validates all tool inputs using the parseArgs helper. parseArgs parses incoming data against the appropriate TypeBox schema and throws an error when the arguments do not match the expected structure.

Permissions & Security

The server implements a comprehensive security model with granular permission controls:

Directory Access Control

Operations are strictly limited to directories specified during startup via args
All operations (including symlink targets) must remain within allowed directories
Path validation ensures no directory traversal or access outside allowed paths

Permission Flags

--readonly: Enforces read-only mode, overriding all other permission flags
--full-access: Enables all operations (create, edit, move, delete)
Individual permission flags (require explicit enabling unless --full-access is set):
- --allow-create: Allow creation of new files and directories
- --allow-edit: Allow modification of existing files
- --allow-move: Allow moving/renaming files and directories
- --allow-delete: Allow deletion of files and directories

Default Behavior: If no permission flags are specified, the server runs in read-only mode. To enable any write operations, you must use either --full-access or specific --allow-* flags.

Symlink Handling

By default, symlinks are followed when both the link and target are within allowed directories
--no-follow-symlinks: treat symlinks as regular files and refuse to traverse their targets, preventing escapes via linked paths

See examples/mcp_permissions.json for sample configurations using these flags.

Build

To compile the project locally run:

bun run build

Run the test suite with:

bun test

Docker build:

docker build -t mcp/filesystem -f Dockerfile .

License

This MCP server is licensed under the MIT License. This means you are free to use, modify, and distribute the software, subject to the terms and conditions of the MIT License. For more details, please see the LICENSE file in the project repository.