MCP server that provides shell command execution and file operations with configurable command whitelisting and audit logging
Six tools with basic schemas and descriptions, but significant gaps in parameter documentation, error guidance, and output structure. Tool names are action-oriented (cmdshell, writeFile, readFile, listFiles, applyPatch, fetchUrl), but descriptions lack depth on when to use each tool and what downstream operations expect. Parameters have types but many lack descriptions (e.g., 'args' in cmdshell and applyPatch are minimally documented). No output schemas documented. Error handling returns raw strings without recovery guidance. Security concerns: writeFile and cmdshell are high-risk WRITE operations with limited validation messaging.
Apply context/unified diff text using patch; args are passed to patch.
Execute a whitelisted shell command with optional arguments
Fetch content from a URL and optionally parse HTML
List entries in a directory beneath the current directory.
Read and return a UTF-8 text file beneath the current directory.
write text into a local file in the current directory
Parameter descriptions are minimal or missing. 'args' in cmdshell and applyPatch lack detail on format, constraints, or examples. 'context' in applyPatch is underdocumented (what is 'fuzz'?). LLMs cannot infer valid input without explicit guidance.
No output schemas documented. Tools return plain strings; LLMs cannot parse structured results or extract IDs for chaining. E.g., listFiles returns newline-separated names but no metadata; readFile returns raw text without encoding/size info.
Error messages are generic strings without recovery guidance. 'Error: Command execution timed out' or 'Error writing file' do not tell the LLM what to do next (retry? ask user? try alternative tool?). No error classification (retryable vs fatal).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
writeFile and cmdshell are destructive WRITE operations but lack confirmation/dry-run patterns. An agent could accidentally overwrite files or execute unintended commands. No idempotent hints or destructive markers in tool definitions.
cmdshell whitelist is hardcoded and includes risky commands (cp, mv, mkdir, awk). No per-request scope declaration or permission gating. Audit logging exists but tool definitions do not declare required permissions (read:filesystem, write:filesystem, execute:shell).