Loading ToolBench
MCP server for controlling Windows system media, audio, power, brightness, processes, and other system functions
24 tools with complete schemas and descriptions, but significant quality gaps. All tools have descriptions (10-200 chars, mostly adequate) and input schemas with proper types. However, many descriptions lack actionable context for LLM selection. No output schemas documented. Error handling is absent, tools do not guide recovery or categorize failures. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite 6 DESTRUCTIVE/IRREVERSIBLE tools. Parameter descriptions are present but often generic. No pagination support for list-returning tools. Security concerns: send_keystrokes and launch_app accept arbitrary input without sanitization; no permission gates on destructive operations.
Force close a running Windows process by name or PID (e.g. 'notepad' or '1244').
Empty the Windows Recycle Bin.
List available system audio output and input hardware devices.
Get battery status, remaining power percentage, charging/AC state, and estimated runtime (supported on laptops).
Get the current screen brightness level (0 to 100) if supported (typically laptops).
Get the current text contents from the Windows clipboard.
Get space metrics (total, used, free) for all active storage drives.
No output schemas documented. LLMs cannot predict response structure, forcing them to guess field names and types for downstream tool calls. Breaks tool chaining.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Six tools marked DESTRUCTIVE/IRREVERSIBLE (system_power_control, close_process, empty_recycle_bin, send_keystrokes, launch_app, set_clipboard) lack explicit hints. LLMs cannot distinguish safe reads from dangerous writes.
No error handling or recovery guidance. Tools do not return actionable error messages. E.g., if launch_app fails, LLM receives no guidance on retry strategy or alternatives.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Get graphics card (GPU) details, driver version, memory, and status.
Get local network IP address, network adapter name, and external IP (if online).
Get the current system resource usage status including average CPU load percentage, RAM (total, used, free, and percentage), C: Drive disk usage (total, used, free, and percentage), and system uptime.
Get the top 5 running processes consuming the most CPU percentage.
Get the current master audio volume level (0 to 100) and mute status.
Scan and list nearby Wi-Fi network SSID names and signal strength.
Launch a Windows application or command (e.g. 'notepad', 'calc', 'explorer').
Simulate media keys playback controls for applications like Spotify, YouTube (Chrome), or media players.
Open a specified web URL in the default or a specified custom web browser.
Send keyboard keystrokes to the currently active foreground window to interact with web pages or apps (e.g. typing text, pressing Enter/Tab).
Set the screen brightness level (0 to 100) if supported.
Write new text to the Windows clipboard.
Mute or unmute the system master audio.
Set the system master audio volume level.
Minimize all active GUI windows to show the Windows desktop.
Control Windows system power options (lock screen, sleep/suspend, schedule shutdown, schedule restart, abort shutdown).
Capture a PNG screenshot of the primary display screen and save it locally.
send_keystrokes and launch_app accept arbitrary strings without input validation or sanitization. Vulnerable to command injection and prompt injection attacks. No constraints on keystroke sequences or app names.
No permission gates on destructive operations. system_power_control (shutdown/restart), close_process, and empty_recycle_bin execute without authorization checks. No audit trail or scope declarations.