NightMend MCP Server has critical definition quality issues. Only 1 tool is exposed (get_servers_health), and while it has an input schema and description, the schema lacks proper type definitions and the overall tool ecosystem is minimal. The input schema shows parameter names and descriptions but is missing explicit 'type' declarations in the JSON Schema format. The tool description is generic (24 chars) and does not explain WHEN to use this tool, what it returns structurally, or dependency relationships. No output schema is documented. The tool itself is read-only and low-risk, but the definition quality falls well below production baseline (baseline avg tool description: 194 chars; here: 24 chars). No error handling patterns are visible, no parameter constraints (e.g., limit bounds, enum for status_filter), and no guidance on pagination or large result handling.

• Expand tool description from 24 to 150 - 200 characters following pattern:tool-description, include: what the tool does, when to use it (monitoring dashboard refresh?), what data it returns (server name, status, uptime%, metrics), and any prerequisites (e.g., 'requires connected agents')
• Add explicit 'type' field to each parameter in the input schema: limit should be type='integer', status_filter should be type='string'
• Replace status_filter free-form string with an enum constraint: {"enum": ["online", "offline", "warning"]}, this self-documents valid options and prevents LLM hallucination
• Add explicit bounds to limit parameter: {"type": "integer", "minimum": 1, "maximum": 100, "description": "Max servers returned (1 - 100, default 10)"}, prevents invalid values and documents the ceiling
• Document the output schema explicitly in the tool description or as a separate field, LLM must know returned fields (e.g., server_id, server_name, status, uptime_percent, last_check_time, metrics{cpu, memory, disk})
• Clarify pagination behavior, does calling with limit=10 return 10 items or 10+? Is there a next_cursor? Can the same limit be used to fetch the next page, or do we need an offset/page parameter?
• Add error handling guidance to the description: 'Returns 400 if limit < 1 or > 100. Returns 400 if status_filter not in [online, offline, warning]. Returns 503 if health check service unavailable, retry after 5s.'
• Consider splitting into discovery tool (list_servers with pagination) and metric tool (get_server_metrics with server_id) if the current tool masks two distinct responsibilities