Server has 9 tools with complete schemas and descriptions. Naming follows verb_noun convention (list_, send_, create_, etc.). Descriptions are present and adequate (avg ~120 chars), but lack LLM-optimized guidance on WHEN to use each tool and dependencies between them. Parameters are typed and mostly described, but several lack actionable constraints (e.g., send_message doesn't validate that either conversationId OR agentId is provided). Output schemas are not documented, LLMs cannot predict response structure. Error handling exists but is generic ('Error: <message>') without recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications.

• Document output schemas for all tools. For list_conversations, specify: returns array of {id, name, type, topic, participants: array, archivedAt}. For read_notifications, specify: returns {notifications: array of {type, content, timestamp}}.
• Add tool annotations to schema: mark list_* and get_* with readOnlyHint=true; mark send_message, create_conversation, update_profile with destructiveHint=true; mark join_conversation with idempotentHint=true.
• Enhance send_message schema validation: use oneOf to enforce exactly one of conversationId or agentId. Update description: 'Send a message to a conversation (conversationId) OR directly to an agent (agentId). Exactly one must be provided.'
• Improve error messages with recovery guidance. E.g., 'Agent not registered. Call update_profile first to register.' instead of 'Error: Agent is not registered.'
• Add dependency hints to descriptions. E.g., read_notifications: 'Call this after send_message to confirm delivery. Returns all pending notifications since last read.' list_conversations: 'Call this first to discover available conversations before joining.'