Loading ToolBench
MCP server + local REPL agent (Strands SDK): identity, cross-agent memory, DMs, tool forge, scheduled jobs, local shell/file/http tools, available to any MCP client or straight from your terminal.
This MCP server exposes 7 Android device control tools via STDIO. Critical gaps: 5 of 7 tools lack visible input schemas entirely (set_brightness, play_sound, schedule_alert, cancel_alerts, vibrate has partial schema). Descriptions are present but generic (10-50 chars), below the 50-200 char LLM-optimized baseline. No parameter descriptions for most tools. No output schemas documented. No error handling guidance. Tool names are action-verb based (good), but lack clarity on what parameters they accept or what they return. The vibrate tool shows a complex Kotlin implementation with waveform logic, but the MCP schema is incomplete. No evidence of permission gates, audit trails, or input validation at the MCP layer.
Cancel scheduled alert notifications
Copy text to the user's clipboard with validation and truncation to 10,000 characters
Open a URL on the device with scheme allowlist validation (https, http, geo, spotify, music, mailto)
Play a sound on the device, respecting quiet hours
Schedule an alert notification on the device
Set device screen brightness
Vibrate the device with named patterns (double, success, warning, error, heartbeat, sos, long, tap, escalate, wave) with configurable repetitions and intensity
5 of 7 tools have no visible input schema (set_brightness, play_sound, schedule_alert, cancel_alerts missing entirely).
Descriptions are 10-50 characters (e.g., 'Set device screen brightness'), below the 50-200 char LLM-optimized baseline. Too short to guide LLM tool selection or explain when to use each tool.
No parameter descriptions for set_brightness, play_sound, schedule_alert, cancel_alerts. LLMs cannot infer what parameters control without explicit descriptions.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract return values without knowing the response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
No error handling guidance. Tools that modify device state (vibrate, play_sound, schedule_alert) should document failure modes and recovery steps (e.g., 'If quiet hours are active, play_sound will fail, check device settings').