An MCP server that executes shell commands on Linux systems with privilege separation and session management
Server defines 6 tools with explicit schemas and descriptions. Naming follows verb_noun convention (execute_command, read_output, list_sessions). Descriptions are present (avg ~80 chars) but lack LLM-optimized guidance on WHEN to use each tool or dependencies between them. All tools have typed input schemas with parameter descriptions. However, output schemas are not documented, callers must infer structure from code. Error handling is minimal; no recovery guidance provided. Security concerns: execute_command accepts sudo flag and arbitrary shell commands with minimal validation, creating injection risk.
Close a shell session and terminate its process
Execute a shell command in a persistent session with output streaming and timeout handling
Read the last N characters of command output
Send input to an interactive command waiting for user input
List all active shell sessions
Read output from a command by character offset range
No output schemas documented. Callers cannot predict response structure (OutputSlice fields: text, start, end, total, discarded). LLMs must infer from code or trial-and-error.
execute_command accepts arbitrary shell commands with sudo flag. No input validation, sanitization, or command injection guards. Descriptions do not warn of destructive capability or require confirmation.
Descriptions lack WHEN-to-use guidance and tool dependencies. E.g., execute_command does not explain that output is streamed and must be read via read_output or get_output_tail. No dependency hints.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | <=2025-11-25 | v2 |
Error handling is absent. No error responses guide LLM recovery (e.g., 'session not found, call list_sessions() first'). No categorization of retryable vs fatal errors.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe reads from destructive writes without parsing descriptions.