MCP server for accessing Garmin Connect health data in Claude Desktop
Server has 9 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-100 chars, mostly adequate) and input schemas with type definitions. However, parameter descriptions are sparse or missing context. No output schemas documented. Error handling is minimal, Python script failures surface raw exceptions rather than actionable recovery guidance. Tools are well-named with clear verbs (get_*, generate_*). No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. The generate_chart tool is WRITE-class but lacks confirmation/dry-run pattern. Parameter validation rules (e.g., date format, day ranges) are stated in descriptions but not enforced via JSON Schema constraints (minValue, maxValue, pattern). No pagination support despite potential for large result sets (e.g., get_activities over 30+ days). Response structure is inferred from Python backend but not formally documented in tool definitions.
Generate an interactive HTML chart (sleep, body_battery, hrv, activities, or dashboard)
Get workout/activity data including type, duration, calories, and distance
Get Body Battery data (Garmin's recovery metric, 0-100)
Get resting, max, and min heart rate data
Get Heart Rate Variability (HRV) data in milliseconds
Get sleep data including hours, stages (light/deep/REM), and quality scores
No output schemas documented. LLMs cannot predict response structure, forcing them to guess field names and types for downstream reasoning. Baseline: 100% of A+ tools document return types.
Destructive tool (generate_chart, WRITE risk) lacks confirmation/dry-run pattern. Agents can overwrite files without user approval. No idempotentHint or destructiveHint annotations.
Error handling delegates to raw Python exceptions. No recovery guidance, categorization (retryable vs fatal), or actionable error messages. E.g., 'Python script failed: ...' tells LLM nothing about next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Get all-day stress level data
Get combined health summary with averages across all metrics
Get user profile information and connected devices
Parameter constraints (date format YYYY-MM-DD, day ranges 1 - 365) stated in descriptions but not enforced via JSON Schema (minValue, maxValue, pattern). LLMs cannot parse prose constraints reliably.
No pagination support. Tools returning lists (get_activities, get_sleep_data) lack limit/offset and total_count. Large result sets risk context window exhaustion.