Four read-only tools with complete input schemas and descriptions. All tools have proper verb_noun naming (brain_search, brain_get, brain_list_by_date, brain_list_by_label). Descriptions are adequate (50-100 chars) but lack actionable context about when to use each tool vs. alternatives. Parameters are well-typed with descriptions, but lack enum constraints for answer_mode and match_all. Output schemas are not documented, LLMs cannot predict response structure. Tool annotations present (readOnlyHint, idempotentHint, openWorldHint) but error handling and recovery guidance missing.

• Document output schemas for all tools. Example: brain_search returns {results: [{note_id, title, content, confidence, source_refs}], total_count, next_cursor}. This enables LLMs to chain results into downstream calls.
• Add enum constraint to answer_mode parameter: enum=['conservative', 'exploration']. Prevents hallucinated values.
• Enhance tool descriptions with selection guidance. E.g., brain_search: 'Use for keyword-based discovery across all notes. Use brain_list_by_date for temporal queries, brain_list_by_label for tag-based filtering.'
• Add error recovery hints to descriptions. E.g., brain_get: 'Returns note by ID. If note_id is unknown, call brain_search() first to discover available notes.'
• Return total_count and next_cursor in paginated responses (brain_list_by_date, brain_list_by_label) so LLMs know when to fetch additional pages.