Server has 8 tools with explicit schemas and descriptions. Naming follows verb_noun convention (list_, prepare_, commit_, write_, browse_, inspect_, read_). Descriptions are present but brief (avg ~80 chars). Schemas use Zod validation with type constraints. Critical gaps: parameter descriptions are minimal or missing context; output schemas are not documented; error handling lacks recovery guidance; no pagination support for browse_node despite potentially large result sets.

• Document output schema for each tool. Example: 'list_controls returns {controls: [{name: string, available: boolean, riskLevel: string}], total: number}'.
• Expand parameter descriptions with format and constraint details. For 'value' param: 'The new value for the control. Format depends on control type (string, number, boolean, or enum). See list_controls for valid types per control.'
• Add pagination guidance to browse_node description: 'Results are paginated (default 50 per page). Use continuation token from previous response to fetch next page. Returns total count and next_cursor if more results exist.'
• Implement error responses with recovery hints. Example: 'Control not found. Call list_controls to see available controls.' or 'Permission denied. This control requires operator approval.'
• Add dry-run support to write_control: accept optional 'dryRun: boolean' parameter. Return what would happen without executing, letting LLM preview before committing.