Teak provides 20 tools with mostly complete schemas and descriptions. The MCP tools (teak_v1_*) have well-structured JSON schemas with proper type definitions, enums, and parameter descriptions. However, 5 tools (search-cards, get-card, get-recent, list-tags, save-card, favorite-card) are inferred from package.json metadata only, no explicit schema or implementation visible in source. Descriptions are generally adequate (50-150 chars) but lack actionable context for LLM selection. Error handling is minimal; no recovery guidance or categorization. Tool naming is mostly clear (verb_noun pattern) but some tools have overlapping responsibilities (e.g., teak_v1_list_cards vs teak_v1_search_cards both retrieve cards). Output schemas are not documented in the source.

• Document output schemas for all tools. Example: teak_v1_list_cards should return {cards: [{id, title, type, tags, ...}], nextCursor, total}. This enables LLMs to chain calls and extract required IDs.
• Add error recovery guidance to descriptions. E.g., teak_v1_get_card: 'If card not found, try teak_v1_search_cards() with a partial title or tag filter.'
• Consolidate list_cards, search_cards, and list_favorite_cards into a single search_cards tool with optional filters (favorited: boolean, type: enum, tags: array). Reduces LLM decision overhead.
• For destructive operations, add a dry_run parameter or a separate preview_delete tool that shows what will be deleted without executing. Prevents accidental data loss.
• Add parameter descriptions to Raycast tools (search-cards, get-card, etc.) and expose their input schemas in the MCP server definition, not just package.json metadata.