Loading ToolBench
An MCP server for controlling Windows system volume, audio devices, and sound profiles with context-aware monitoring
This server has significant definition quality gaps. Tool descriptions are present but minimal (10-30 chars), well below the 50-200 char baseline for LLM-optimized descriptions. Input schemas are visible in the rubric summary but the actual source code shows a custom HTTP handler with command-based routing, not explicit MCP tool registration. Parameter descriptions are sparse or missing. The 'help_prompt' tool is a meta-tool that doesn't fit standard patterns. Error handling is basic (success/message fields) without recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk distinctions.
Get the current mute status.
Get the current master volume level.
Create a help prompt with available commands.
List available audio devices.
Mute the system audio.
Set the master volume level.
Unmute the system audio.
Tool descriptions are critically short (10-30 chars), far below the 50-200 char baseline. LLMs cannot determine when/why to select tools. E.g., 'Get the current master volume level' lacks context on use cases, return format, or when to call vs alternatives.
No input parameter descriptions visible in source code. The 'level' param for set_volume has a description in the rubric summary, but 'device_type' for list_audio_devices lacks clarity on valid values ('output' vs 'input'). LLMs cannot infer parameter meaning from names alone.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk distinctions. set_volume, mute_audio, unmute_audio are marked WRITE but lack destructiveHint. This prevents agents from understanding retry safety and side-effect scope.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | <=2025-11-25 | v2 |
Error responses lack recovery guidance. Source shows generic {'success': False, 'message': '...'} responses. No actionable next steps for LLMs (e.g., 'Failed to set volume. Ensure level is 0.0 - 1.0'). Agents cannot self-correct.
'help_prompt' is a meta-tool that returns help text, not a domain action. This violates single-responsibility principle and wastes a tool slot. Remove it; let the LLM infer available tools from descriptions.