All 5 tools have zero-parameter schemas with minimal descriptions (10-30 chars). No input validation, no parameter guidance, no output schema documentation. Tool names lack action verbs (health.check, identity.profile, auth.session use dot notation rather than verb_noun). Descriptions are too brief to guide LLM selection. No error handling guidance. Tool annotations present but insufficient to compensate for missing core definition quality.

• Rename tools to verb_noun convention: health.check → check_health, identity.profile → get_profile, auth.session → get_auth_session. This enables LLMs to parse intent from the name alone.
• Expand all descriptions to 50-200 characters, explaining WHAT the tool does, WHEN to call it, and WHAT it returns. Example: 'get_profile: Retrieve the authenticated user's profile including name, email, and roles. Call this first to verify the current user's identity and permissions.'
• Document output schemas for all tools. Specify the structure of returned data (e.g., { status: string, uptime_ms: number, version: string } for check_health). This enables agents to extract and chain results.
• Add error handling guidance to auth.session and identity.profile descriptions: 'If the OAuth token is missing or expired, the tool returns a 401 error. Call authenticate_user() to obtain a fresh token, then retry.'
• Clarify the distinction between .profile and .profile_card tools. If .profile returns structured data and .profile_card renders UI, state this explicitly: 'get_profile returns structured data; display_profile_card renders an interactive UI card. Use get_profile for data extraction, display_profile_card for user-facing output.'