Loading ToolBench
Model Context Protocol server for Nightscout — lets LLMs read glucose, treatments, and analytics from a personal Nightscout instance.
13 tools with consistent naming (verb_noun pattern: get_, detect_, calculate patterns). Descriptions are present and domain-specific (50-150 chars, within baseline 34-392). However, input schemas are minimal: most tools accept only 'days' or 'hours' integers with no min/max bounds, and several accept empty objects. Output schemas are completely undocumented, LLMs cannot predict return structure. Error handling is absent from visible code. Parameters lack constraint documentation (e.g., 'days' has no stated range). Tool composition is sound (each does one thing), but schema completeness and output documentation are major gaps.
Retrieve the change history of a single AAPS setting over a time window.
Retrieve all AAPS settings changes between two timestamps.
Retrieve the most recent AAPS settings snapshot from the local ingest database.
Generate a comprehensive markdown report suitable for sharing with a healthcare provider.
Audit glucose readings against consensus targets and report unmet targets.
Assess whether the available data meets AGP (Ambulatory Glucose Profile) consensus standards for analysis.
Identify recurring glucose patterns (dawn phenomenon, post-meal spikes, overnight lows, etc.) over a time window.
Output schemas completely undocumented. LLMs cannot predict return structure (fields, types, nesting). Tools return complex analytics (glucose stats, patterns, risk indices) but no schema visible in code.
Input parameter constraints missing. 'days' and 'hours' parameters lack min/max bounds, default values, or validation guidance. LLMs may pass invalid values (0, negative, >365).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | <=2025-11-25 | v2 |
Fetch the most recent SGV (sensor glucose value) and trend arrow.
Compute glucose statistics (mean, TIR, GMI, etc.) over a time window.
Calculate the Glycemia Risk Index (GRI) — a composite risk metric combining hyperglycemia and hypoglycemia excursions.
Probe Nightscout connectivity and return server status.
Count and characterize low blood glucose episodes (Level 1 <70 mg/dL, Level 2 <54 mg/dL) over a time window.
Derive effective insulin sensitivity factor (ISF) from correction boluses and their glucose outcomes.
No error handling or recovery guidance visible. Tools do not document what happens on invalid input, missing data, or API failures. LLMs receive no actionable error messages.
Empty input schemas for 3 tools (health_check, get_current_glucose, aaps_settings_snapshot).
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are READ_ONLY but this is not declared in schema. Agents cannot infer safety properties.