A comprehensive Model Context Protocol (MCP) server for file system operations with advanced features like file watching, search, and batch operations.
This server defines 16 filesystem tools with complete input schemas and clear descriptions. However, it lacks output schema documentation, has minimal error handling guidance, and missing security annotations. Tool naming follows verb_noun pattern (read_file, write_file, etc.) which is good. Parameter descriptions are comprehensive with type constraints (enums for encoding, etc.). The server does NOT implement tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite having 5+ tools with clear side effects. Error handling exists in code but lacks guidance in tool descriptions. Security considerations (path traversal, input validation) are not documented in tool descriptions despite being critical for filesystem ops.
Create archive from files
Compare two files and identify differences
Copy file from source to destination
Create directory with optional recursive creation
Delete file or directory
Extract archive to destination directory
Find files matching a pattern
No output schemas documented. Tool descriptions state what the tool does but LLMs cannot infer the response structure. For example, get_file_info, list_directory, and find_files have no documented output format, agents must guess what fields are returned.
Missing tool annotations. Destructive tools (delete_file, write_file, move_file, archive_files, extract_archive) lack destructiveHint=true annotation. Read-only tools (read_file, get_file_info, list_directory, etc.) lack readOnlyHint=true. Idempotent operations (copy_file with overwrite, read_file) lack idempotentHint. These annotations are part of the current MCP spec and guide agent safety planning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Get total size of directory recursively
Get detailed file information including metadata
List directory contents with optional filtering and recursion
Move file from source to destination
Read file content with optional encoding and range
Search for text patterns in files
Stop watching a file or directory
Watch file or directory for changes
Write content to file with optional encoding and directory creation
No error recovery guidance in tool descriptions. Descriptions do not explain what happens on common failures (file not found, permission denied, path traversal blocked, disk full). For example, delete_file does not say 'If path does not exist and force=false, an error is returned' or provide recovery steps. Agents have no guidance on retry strategies.
Security constraints not documented in tool descriptions. Filesystem tools are vulnerable to path traversal attacks (e.g. '../../../etc/passwd'). No tool description mentions input validation, sandboxing, or root path restrictions. The policy.ts file exists but policy constraints are not visible in tool descriptions, agents cannot know what paths are allowed.
Result pagination not documented. list_directory, find_files, and search_in_files lack pagination parameters (limit, offset/cursor, page) and response metadata (total_count, has_more, next_cursor). Large directory listings or search results could blow context windows. No cap on result size mentioned in descriptions.
watch_file/stop_watching lack stateful operation clarity. Tools register file watchers but do not document how results are delivered (via resources? streaming? polling?). The watch_file description does not explain how an agent receives watch notifications or what the expected workflow is.
Default values on destructive parameters. write_file defaults 'append=false' (overwrite), delete_file defaults 'recursive=false' and 'force=false'. While these defaults are conservative, they should be explicitly highlighted in descriptions to prevent accidental data loss if an LLM omits the parameter.