WhisperGraph MCP demonstrates solid tool design with consistent naming patterns (verb_noun), comprehensive parameter descriptions, and proper input validation. All 8 tools have clear descriptions (100-200 chars) and typed schemas. However, output schemas are not explicitly documented in the source, and error handling, while present, lacks structured recovery guidance. Tool composition is clean with single responsibilities. Parameter validation is defensive (regex whitelisting for indicators, labels, variant names) and returns structured error objects rather than throwing, which is excellent for agent resilience.

• Document output schemas for all tools. E.g., explain_indicator returns {rows: [{indicator, available, error?, explanation, ...threatData}]}. Include field types and when fields are present.
• Enhance error messages with actionable recovery steps. Change 'invalid_indicator' to 'Invalid indicator format. Expected: IPv4 (e.g., 8.8.8.8), IPv6, hostname, CIDR (e.g., 8.8.8.0/24), or ASN (e.g., AS15169).'
• Add tool selection guidance in descriptions. For query: 'Use for custom Cypher exploration. For threat assessment, prefer explain_indicator. For domain analysis, prefer domain_variants.'
• For run_recipe, document expected input schema per recipe type. Either enumerate common recipes with their input shapes, or link to a recipe catalog endpoint.
• Add pagination guidance to list_labels and list_recipes if they can return large result sets. Specify max result count and whether pagination is supported.

Inferred effective spec: 2026-07-28+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry