Server has 19 tools with mostly complete schemas and descriptions. Naming follows verb_noun convention well (ingest_document, search_knowledge, create_entity). Descriptions average ~100-150 chars, meeting baseline. However, several critical gaps: (1) No output schemas documented for any tool, LLMs cannot plan downstream calls or extract fields. (2) Many parameters lack type constraints (e.g., entity_type, relationship_type, direction accept free-form strings with no enum). (3) Error handling is absent, no recovery guidance, no actionable error messages. (4) Destructive tools (prune_stale, archive_entity) lack confirmation/dry-run patterns. (5) Parameters like 'certainty' and 'priority' should be enums but are described as free strings. Strengths: clear naming, reasonable description lengths, parameter descriptions present for most tools.

• Document output schemas for all 19 tools. Specify field names, types, and whether results are paginated. Example: search_knowledge returns {results: [{entity_id, name, score, type}], total_count, next_cursor}.
• Convert free-form string parameters to enums: entity_type → enum [person, concept, event, location, organization]; direction → enum [incoming, outgoing, both]; certainty → enum [confirmed, likely, speculative]; priority → enum [low, medium, high]; order → enum [ASC, DESC].
• Add dry_run parameter to prune_stale (already present) and archive_entity. Document that dry_run=true returns what would be deleted without executing.
• Add error handling documentation to each tool. Example: 'Returns 404 if entity_id not found. Try search_knowledge() to find the entity first. Returns 400 if entity_type is invalid, must be one of: person, concept, event, location, organization.'
• Standardize pagination across all list/search tools: add limit (default 20, max 100) and offset (default 0) parameters; return total_count and next_cursor in responses.

Inferred effective spec: <=2025-11-25.

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