A collection of MCP servers including exec (command execution), filesystem (file operations), and web (web scraping/content retrieval)
This collection exposes 14 destructive and irreversible tools (command execution, process termination, package installation) with minimal safety guardrails. Tool names are action-oriented (run_command_sync, execute_command, kill_process) but descriptions are terse (avg ~60 chars, well below the 194-char baseline). Most critically: NO input schemas are visible in the provided source code, only parameter names and types are listed informally. Parameters lack enum constraints, format specifications, and validation guidance. Error handling is absent, no recovery guidance, no categorization of retryable vs. fatal errors. No confirmation/dry-run patterns for irreversible operations. Security is severely compromised: shell execution accepts arbitrary commands with no injection protection, no permission gates, no audit logging visible. The exec.py file is not shown in full, so tool registration details cannot be verified; this forces a conservative cap on per-tool scores.
Prevent specific commands from being executed
Install multiple packages across different managers
Run shell commands asynchronously with configurable wait time. The process doesn't terminate after wait time, but is run in a session pool that can be accessed later.
Run multi-line code in Python, JavaScript, Rust, Go, or Bash asynchronously
Kill a specific running session
Install specific packages or tools
Terminate a process by its PID
No input schemas visible in source code. All 14 tools scored 0 for schema, cannot verify JSON Schema definitions, type constraints, or parameter validation.
Destructive operations (execute_command, kill_process, force_terminate, install_command) lack confirmation/dry-run patterns. No error guidance for recovery. Agents can irreversibly delete processes or install packages without safeguards.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 29 | 2026-07-28+ | v2 |
View all commands on the blacklist
Show available tools and commands
List all running system processes by PID
View all active command sessions
Read output from a running session
Run a shell command synchronously with timeout and return the result.
Allow previously blocked commands
Command execution tools (run_command_sync, execute_command, execute_script) accept arbitrary shell commands with no injection protection, no allowlist, no sanitization visible. LLM can be tricked into passing malicious payloads (e.g., 'rm -rf /'). Critical security gap.
Descriptions are terse (avg ~60 chars, baseline 194 chars). No guidance on WHEN to use each tool, WHAT it returns, or HOW to handle errors.
No audit logging, permission gates, or scope declarations visible. Tools that kill processes or install packages should require explicit authorization and log all invocations for compliance. No evidence of least-privilege design.