12 tools with complete schemas and descriptions, but significant gaps in parameter documentation and error handling. Most tool descriptions are adequate (50-150 chars) but lack actionable context for LLM selection. Parameters like 'type', 'tag', 'source' lack descriptions entirely. No error recovery guidance, no output schema documentation, no tool annotations. Naming is consistent (brain_* prefix) but generic verbs (brain_put, brain_get) don't clearly distinguish intent. No idempotency hints for write operations.

• Add parameter descriptions to all undescribed fields: 'type' (enum: person|company|deal|...), 'source' (enrichment source name), 'data' (JSON object structure). Reference pattern:tool-description.
• Document output schemas for all tools. E.g. brain_search returns [{slug, title, type, score}]; brain_stats returns {page_count, link_count, embedding_count}. Enables LLM to plan downstream calls.
• Add error recovery guidance: 'Page not found. Try brain_search() to find the correct slug.' Implement pattern:recovery-guide for all read operations.
• Add tool annotations: mark brain_put, brain_timeline_add, brain_tag, brain_raw with destructiveHint=true and idempotentHint=false. Guides LLM on retry safety.
• Rename brain_get → brain_read_page, brain_put → brain_write_page for clarity. Use verb_noun convention to disambiguate intent from name alone.