This server provides 4 tools with explicit schemas and descriptions, but exhibits significant gaps in parameter documentation, output schema clarity, and error handling guidance. Tool names follow verb_noun convention (social_read, social_update, social_score, social_draft), which is good. However, descriptions lack actionable context for LLM selection, and critical parameters are underdocumented. The social_draft tool exposes LLM integration details that should be abstracted. No error recovery guidance is present. Output schemas are not formally documented in any tool. This is a typical community-grade server with foundational structure but missing production-quality polish.

• Add formal output schema documentation to every tool. Example for social_read: 'Returns {signals: {energy: number 0-1, engagement: number 0-1, mode: string, notes: string}, timestamp: ISO8601}'. This enables LLMs to plan chained calls.
• Refactor social_draft to hide cortex-engine and LLM details behind a clean interface. Move VOICE_PROMPT and ctx.llm.generate logic into a private helper; expose only author, text, score, and return {draft: string, platform: string}.
• Add error handling with recovery guidance. Example: 'If cortex memory unavailable, return {draft: fallback_reply, warning: "Memory unavailable. Reply generated without context."}' so the LLM knows the limitation.
• Document parameter constraints in descriptions. Replace 'namespace (defaults to default)' with 'Namespace for signal storage (string, defaults to "default"). Use to isolate signals by project or context.'
• Add per-parameter defaults and validation. For social_score, state: 'If score omitted, defaults to 50. If engagementScore omitted, defaults to 0. If matchedRule omitted, defaults to "none".'