MCP server for filesystem access with security restrictions and SSE transport
This filesystem MCP server demonstrates solid definition quality with 10 tools covering read, write, and navigation operations. All tools have explicit names, descriptions, and JSON Schema input definitions derived from Zod schemas. Naming follows verb_noun convention (read_file, write_file, search_files). Descriptions are present for all tools (range 50-150 chars) and most parameters. However, there are notable gaps: output schemas are not documented, error handling lacks recovery guidance, and some parameter descriptions could be more precise about constraints and use cases. The server correctly uses Zod for schema validation at runtime, which is good for security, but does not expose detailed error classifications or recovery hints to the LLM.
Create a new directory
Get a tree view of a directory and its contents
Edit a file by replacing specific text patterns. All old_text values must match exactly (including whitespace) or the operation will fail. Use dryRun to preview changes.
Get detailed information about a file or directory
List contents of a directory
Move or rename a file or directory
Read the complete contents of a file from the filesystem
Output schemas are not documented. Tools return structured objects (FileInfo, search results, file contents) but the LLM has no declared specification of what fields to expect. This forces the LLM to infer the response structure and makes downstream tool chaining ambiguous.
Error messages lack recovery guidance. The code validates paths and throws errors like 'Access denied - path outside allowed directories' but does not guide the LLM on what to do next (e.g., 'Try reading from one of these allowed directories: [list]').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 68 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 35 | - | v1 |
Read the complete contents of multiple files
Search for files and directories by name in a directory tree
Write content to a file, creating it if it doesn't exist
Parameter descriptions lack precision on constraints and use cases. For example, 'path' parameters describe the path but do not indicate whether home directory expansion (~) is supported, relative vs absolute paths, or symlink handling. LLMs may attempt invalid paths.
Destructive operations (write_file, edit_file, move_file, create_directory) lack confirmation or dry-run support. edit_file offers dryRun=true, which is good, but write_file and move_file do not. An agent could accidentally overwrite critical files or move data without a safety check.
No pagination or limit controls on read_multiple_files and search_files. If an agent requests 1000 files or searches a large directory tree, the response could exhaust the context window. No cursor, offset, or result limit parameters are exposed.
Tool annotations (readOnlyHint, destructiveHint, idempotentHint) are not used. The server does not declare which tools are read-only, destructive, or idempotent in the tool schema. This forces the LLM to infer intent from names and descriptions.
No batch operations. Common agent workflows loop over multiple files (e.g., read_multiple_files calls read_file N times, or search_files results are processed sequentially). Offer batch variants to reduce token waste and latency.