Five read-only monitoring tools with clear, consistent naming (verb_noun pattern: pm2_status, system_resources, nginx_errors, ssl_expiry, http_health). All tools have descriptions (avg 120 chars, within 10-1024 baseline). Input schemas present for 3/5 tools; 2 tools (pm2_status, system_resources) have empty schemas {}. Parameter descriptions are present and specific (e.g., 'How many trailing lines of the error log to return'). However, output schemas are not documented, LLMs cannot predict response structure. Error handling uses generic fail() wrapper with no recovery guidance. No tool annotations (readOnlyHint, idempotentHint). Composition is sound: each tool does one thing; no multi-step dependencies documented.

• Add explicit input schemas to pm2_status and system_resources. Even if they take no parameters, declare inputSchema: { type: 'object', properties: {}, required: [] } to signal 'no inputs accepted' clearly.
• Document output schemas for all tools. Add a 'Returns' section to each tool description or create a separate schema document. Example: 'Returns array of {name, status, cpu, memoryMb, uptimeMinutes, restarts}.'
• Enhance error messages with recovery guidance. Instead of 'Error: ENOENT', return 'Nginx error log not found at /var/log/nginx/error.log. Check NGINX_ERROR_LOG env var or verify file exists.'
• Add tool annotations: set readOnlyHint: true and idempotentHint: true on all five tools. This signals to agents that these calls are safe to retry without side effects.
• Tighten parameter descriptions with format constraints. For ssl_expiry domain, state: 'Domain name (letters, digits, dots, hyphens; 1-253 chars). Example: example.com'.