MCP server for drone fleet management, providing observer tools for telemetry and status monitoring, and remediation tools for MQTT commands, Kubernetes scaling, and human approval workflows.
Server has 7 tools with basic schemas and Italian descriptions. Critical gaps: (1) Most descriptions are under 50 chars and lack actionable context for LLM selection. (2) Parameter descriptions are minimal or missing entirely, e.g., 'minutes_ago' in get_drones_telemetry lacks range/default clarification. (3) No output schemas documented, LLMs cannot infer what fields to expect from responses. (4) No error handling guidance, tools do not explain recovery paths. (5) Tool naming is acceptable (verb_noun pattern) but descriptions fail to distinguish when to use similar tools (e.g., get_drones_status vs get_drones_telemetry). (6) Security-critical tool send_mqtt_command mentions 'destructive actions blocked' but lacks enum constraints on 'action' parameter, LLM could hallucinate invalid actions. (7) request_human_approval has an enum for action_type but payload is a bare object with no schema, LLM cannot validate structure. Baseline: 94% of A+ tools have full param descriptions; this server has ~30% coverage.
Controlla lo stato delle richieste di approvazione
Legge da Kubernetes lo stato attuale dei droni
Interroga InfluxDB per lo stato dei droni recenti
Interroga InfluxDB per gli ordini in sospeso
Richiede approvazione umana per azioni ad alto impatto (scale > 6 droni, comandi distruttivi)
Modifica le repliche del deployment droni su Kubernetes. Se replicas supera il limite di automazione, blocca e indirizza all'approvazione umana.
No output schemas documented for any tool. LLMs cannot infer response structure, forcing them to guess field names and types. This breaks tool chaining and increases hallucination risk.
Parameter descriptions are minimal or absent. 'minutes_ago' lacks default value, range constraints, and use-case guidance. 'action' in send_mqtt_command is free-form string with no enum, LLM will hallucinate invalid actions like 'destroy' or 'explode'.
Tool descriptions are too short (10-50 chars) and lack context for LLM selection. 'Legge da Kubernetes lo stato attuale dei droni' (Italian, ~40 chars) does not explain WHEN to call this vs get_drones_telemetry, or what fields are returned.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
Invia un comando MQTT. Le azioni distruttive sono bloccate: l'agente deve passare per request_human_approval.
request_human_approval accepts 'payload' as a bare object with no schema. LLM cannot validate structure before sending. Should define strict schema for each action_type (e.g., {replicas: integer} for scale_drone_deployment, {target: string, action: string} for send_mqtt_command).
No error handling guidance. Tools do not explain what to do if Kubernetes is unreachable, InfluxDB query fails, or MQTT broker is down. Agents will retry blindly without recovery context.