tvcontrol exposes 51 tools with minimal schema documentation and inconsistent descriptions. Tool names follow verb_noun convention (good), but parameter schemas are not visible in source code provided. Descriptions vary from adequate (health_check: 47 chars) to vague (chart_manage_indicator: 47 chars). No evidence of input validation, error recovery guidance, or output schema documentation. Critical security issue: ui_evaluate is an IRREVERSIBLE tool that runs arbitrary JavaScript with only an environment variable gate (TV_MCP_ADVANCED=1), violating permission-gate and secret-injection patterns. Tool composition is reasonable (single responsibility per tool), but lack of schema visibility and error handling guidance significantly limits LLM usability.

• Add explicit JSON Schema input definitions to all 51 tools. Include type, description, enum constraints, and min/max for every parameter. Use Zod (already a dependency) to generate schemas from TypeScript types.
• Document output schemas for all tools. Specify which fields are returned, their types, and which IDs/references enable chaining to downstream tools (e.g., chart_get_state should return indicator entity IDs for use in chart_manage_indicator).
• Implement permission gates for ui_evaluate. Require explicit user/agent authorization beyond an environment variable. Add audit logging (who, what, when, result). Consider removing arbitrary JavaScript execution entirely or wrapping it in a confirmation-request pattern.
• Add error recovery guidance to all tool descriptions. Example: 'If chart symbol is invalid, call chart_get_state first to see available symbols.' Include categorization (retryable vs user-fixable vs fatal).
• Add confirmation/dry-run support to destructive tools (alert_delete, watchlist_remove, tab_close, state_delete). Return a confirmation_required response or add a confirm=true parameter to prevent accidental deletions.