GlobiGuard MCP server demonstrates solid definition quality with complete input schemas, clear descriptions, and thoughtful parameter design. All 5 tools have explicit JSON Schema definitions with proper type constraints, enums, and length limits. Descriptions are substantive (100-200 chars typical) and explain the governance/authorization context well. Tool naming follows verb_noun convention (govern_action, get_authorization, check_approval). However, output schemas are not documented, responses are inferred from context but not formally specified. Error handling guidance is minimal; tools lack recovery hints for common failure modes. Parameter descriptions could be more explicit about format constraints and dependencies (e.g., payload vs payloadSummary mutual exclusivity is stated but could be clearer).

• Document output schemas for all 5 tools. Specify the exact fields returned, their types, and whether they are always present or conditional. Example: 'Returns {canExecute: boolean, authorizationId: string, nextStep: string, evidence?: object[]}'.
• Add error classification and recovery hints to tool descriptions. E.g., 'If authorizationId is not found (404), call search_authorizations() with a correlationId or workflowRunId. If the service is unavailable (503), retry after 30 seconds.'
• Clarify ID format constraints in parameter descriptions. E.g., 'authorizationId: UUID v4 format (e.g., 550e8400-e29b-41d4-a716-446655440000). Obtain from globiguard_govern_action response.'
• Use JSON Schema oneOf or allOf to enforce payload vs payloadSummary mutual exclusivity at the schema level, not just in prose.
• Document pagination and result limits. If get_audit_event returns unbounded results, add limit and offset parameters (default limit=20, max=100) and return a total_count field.