Single tool with minimal definition quality. list_applications has a basic description (59 chars) but ZERO input schema, the tool accepts no parameters, yet the schema is not explicitly documented. Output is unstructured plain text (YAML dump), not a typed object. No parameter descriptions exist (N/A). No error handling guidance, errors return raw exception strings. Tool naming is acceptable (verb_noun), but lacks pagination, result limits, and structured output. No tool annotations (readOnlyHint, etc.). This server falls well below production baseline.

• Define explicit input schema with limit (1-100, default 20) and offset (default 0) parameters. Document in tool description: 'Returns paginated list of ArgoCD applications. Limit capped at 100 to prevent context window exhaustion.'
• Change output from plain text to structured JSON: return {applications: [{name, namespace, sync_status, health_status, summary}], total_count, offset, limit}. This enables downstream tool chaining and LLM parsing.
• Add error handling that returns {error: string, category: 'retryable'|'user_fixable'|'fatal', recovery_hint: string}. E.g., 'ARGOCD_ENDPOINT not set. Set environment variable and restart server.'
• Add readOnlyHint annotation to list_applications via @mcp.tool(readOnlyHint=True) to signal safety to agents.
• Document API timeout (30s) and result limit (all apps, no cap) in description. Add note: 'For large deployments (>100 apps), use limit parameter to avoid context window overflow.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry