Self-hosted, multi-device activity timeline collector with an MCP server. Tracks what app you're using on Android / iOS / Windows / macOS in real time.
Five tools with clear, domain-specific names (device_status, device_timeline, device_activity_summary, health_summary, health_records) and solid descriptions (avg 180 chars, within baseline 194). All tools have input schemas with Zod validation and type definitions. However, parameter descriptions are sparse or missing entirely: 'deviceId' and 'limit' lack descriptions across multiple tools; 'date' format is documented in device_timeline but not device_activity_summary. Output schemas are not formally documented, responses are text-formatted summaries rather than structured JSON objects, limiting downstream tool chaining. Error handling is minimal (only HTTP status checks, no recovery guidance). No tool annotations (readOnlyHint, etc.). Composition is sound: each tool has one responsibility, and naming follows verb_noun convention.
Per-app usage totals for a day (default today): screen time and switch counts, sorted by time.
Current real-time state of every tracked device: which app is in the foreground, online/offline, battery.
Chronological activity timeline for a day (default today). Optionally filter by deviceId.
Raw Health Connect samples of one metric type over a recent window (default 24h), newest first.
Health Connect summary over a recent window (default 24h): per-metric totals or latest readings (heart rate, steps, sleep, calories, SpO2, blood pressure, weight, etc.). Synced from the user's phone.
Parameter descriptions missing or incomplete. 'deviceId' and 'limit' appear in 4+ tools with no description text; 'date' format documented only in device_timeline, not device_activity_summary. LLMs cannot infer parameter meaning from names alone.
Output schemas not documented. All tools return text-formatted summaries (via text() helper) rather than structured JSON. Downstream tools cannot extract typed fields (e.g., device names, timestamps, metric values) for chaining. Violates pattern:response-shaper.
Error handling lacks recovery guidance. HTTP errors throw generic 'HTTP {status}' messages with no actionable next steps. No categorization (retryable vs. user-fixable vs. fatal). LLMs receive no guidance on how to proceed.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint metadata. All tools are read-only (safe to retry), but this is not declared to the MCP client, forcing conservative retry logic.
Pagination not implemented. device_timeline and health_records accept 'limit' but no offset/cursor or total_count in response. Large result sets risk context window exhaustion; no guidance on how to fetch next page.