MCP server providing health data from Notify for Xiaomi app backup database stored in Google Drive
This MCP server has moderate definition quality with significant gaps. All 6 tools are explicitly registered with descriptions and visible in src/server.py. Naming follows verb_noun convention (nxk_sync_database, nxk_list_tables, nxk_get_sleep, etc.), which is appropriate. However, descriptions are in Norwegian, parameter schemas are minimal (only 3 of 6 tools declare input parameters), and output schemas are not documented. The server uses Pydantic models (SleepParams, HeartRateParams, ActivityParams, QueryParams) internally but does not expose these schemas in the tool definitions visible in the code. Error handling is present with error_handling module imported, but recovery guidance is generic. The per-tool analysis reveals: tools 1-2 have no input parameters (empty {}), tools 3-5 have 3 parameters each (start_date, end_date, limit) with descriptions, and tool 6 has 1 parameter (sql). None of the tools export their response schema in the visible tool definitions, making it impossible for LLMs to understand expected output structure.
Hent aktivitetsdata inkludert skritt, kalorier og distanse. Returnerer aktivitetsdata fra databasen med valgfri datofiltrering. Tidsstempler konverteres til YYYY-MM-DD HH:MM format.
Hent pulsmålinger med lesbare tidsstempler. Returnerer pulsdata fra databasen med valgfri datofiltrering. Tidsstempler konverteres til YYYY-MM-DD HH:MM format.
Hent søvndata med lesbare tidsstempler. Returnerer søvndata fra databasen med valgfri datofiltrering. Tidsstempler konverteres til YYYY-MM-DD HH:MM format.
List alle tabeller i databasen med struktur og antall rader. Viser tabellnavn, kolonner med datatyper, og totalt antall rader for hver tabell i backup-databasen.
Utfør egendefinerte SELECT-spørringer mot databasen. Tillater kun SELECT-spørringer for å beskytte databasen mot skriveoperas joner. Alle andre SQL-kommandoer avvises.
Synkroniser backup-databasen fra Google Drive. Laster ned den nyeste backup.db-filen fra "Notify for xiaomi"-mappen i Google Drive til lokal katalog (~/.notify-xiaomi-mcp/).
Output schemas are not documented. Tools return dicts with arbitrary structure but LLMs cannot see what fields to expect. Tool 1 returns {message, file_size_bytes, download_timestamp}, tools 3-5 return {records, count}, tool 6 returns {rows, column_names}, none of this is declared in the tool definitions.
Input schemas are incomplete for data retrieval tools (3-5). All parameters are documented in docstrings but FastMCP tool decorators do not show explicit JSON Schema with type constraints. The 'limit' parameter should enforce 1-1000 bounds but no validation constraint is visible in the schema.
Descriptions are in Norwegian, not English. While technically valid, international developer teams and LLM multi-lingual models rely on English tool descriptions for optimal performance. This creates a barrier to adoption and may degrade LLM tool selection quality in multilingual deployments.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
nxk_query accepts raw SQL string with no format validation or injection prevention visible in the schema. While the docstring states 'Tillater kun SELECT-spørringer', the tool parameter is a bare string type with no regex pattern or documented validation. The description needs to explicitly state: 'SELECT queries only. INSERT, UPDATE, DELETE, and DROP are rejected.'
Parameters 'start_date' and 'end_date' (in tools 3-5) have descriptions in Norwegian ('Startdato i YYYY-MM-DD format (valgfri)') and are marked optional, but no default values are shown. If an LLM omits both, does it return all records? The description should clarify: 'If omitted, defaults to last 30 days' or 'If both omitted, returns all available data (limited by limit parameter).'
Error handling exists but recovery guidance is minimal. The create_error_response() function is called with categories like 'auth', 'resource', 'execution', but error messages are not shown in the code excerpt. Without seeing error response examples, it is unclear whether LLMs receive actionable guidance like 'User not found. Try calling sync_database first to refresh credentials.'