An AI assistant server that provides tools for time, weather, web browsing, and note management. Integrates with Ollama for LLM capabilities and exposes both MCP and FastAPI interfaces.
Jarvis exhibits significant quality gaps across naming, descriptions, and schema documentation. Tool names lack action verbs (ora_corrente, meteo, apri_sito_web use Italian but lack verb-noun clarity for LLM parsing). Descriptions are present but minimal (10-60 chars, below the 50-200 char LLM-optimized baseline). Parameter descriptions exist but are sparse. No output schemas are documented. The server uses FastMCP decorators (@mcp.tool()) but provides no structured output documentation or error handling guidance. Risk classifications (READ_ONLY vs WRITE) are noted but not reflected in tool annotations (destructiveHint/idempotentHint). No pagination, no error recovery patterns, no input validation guidance.
Aggiunge del testo in fondo ad una nota esistente, senza cancellare il contenuto precedente
Apre un sito web nel browser dell'utente
Legge e restituisce il contenuto di una nota salvata in precedenza
Restituisce il meteo corrente di una città specificata
Crea una nuova nota con un titolo e un contenuto, sovrascrivendo qualsiasi nota con lo stesso titolo
Restituisce l'ora corrente del sistema
Tool names lack action verbs and LLM-parseable intent. 'ora_corrente', 'meteo', 'apri_sito_web' do not follow verb_noun convention (get_current_time, get_weather, open_website). LLMs struggle to infer intent from non-English names without explicit verb prefixes.
Descriptions are too short (10-60 chars vs 50-200 char baseline). 'Restituisce l'ora corrente del sistema' (36 chars) lacks context for LLM selection. No WHEN to use, no prerequisites, no return value documentation.
No output schemas documented. Tools return strings but LLMs cannot infer structure (is meteo returning JSON, plain text, or structured weather data?). No pagination, no field documentation, no chaining IDs for downstream tools.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | <=2025-11-25 | v2 |
No error handling guidance. Tools return strings but no error classification (retryable vs fatal). No recovery hints. E.g., if meteo fails for unknown city, LLM has no guidance to search_cities() first.
Tool annotations missing. Risk classifications (WRITE vs READ_ONLY) are noted in metadata but not exposed as destructiveHint/idempotentHint in MCP schema. LLMs cannot infer side effects from tool definitions alone.