Three well-named, verb-led tools with complete input schemas and clear descriptions. All parameters are typed and described. Tool names (search_tarot_cards, get_tarot_card_love_meaning, draw_tarot_cards) follow verb_noun convention and are unambiguous. Descriptions range 120 - 180 chars, within the 10 - 1024 baseline. All 3 tools have documented input schemas with proper JSON Schema structure. Parameters include enums (language), constraints (count: 1 - 10), and clear descriptions. Error handling is present (invalid slug, empty query) with actionable guidance. Main gaps: output schemas are not formally documented in the tool definitions (only visible in code comments), and no explicit per-tool permission/scope declarations. Composition is clean, each tool has one responsibility, and search_tarot_cards output (slugs) chains directly into get_tarot_card_love_meaning input.

• Add an outputSchema field to each tool definition documenting the return structure. For search_tarot_cards: {type: 'object', properties: {results: {type: 'array', items: {slug: string, name: string}}}}. For get_tarot_card_love_meaning: {type: 'string', description: 'Markdown-formatted card interpretation'}. For draw_tarot_cards: {type: 'string', description: 'Markdown-formatted spread with card names, meanings, and yes/no verdicts'}.
• Add a 'readOnlyHint: true' annotation to all three tools in the tool definition to signal to agents that these are safe, non-destructive operations.
• Expand draw_tarot_cards description to explicitly state: 'Cards are drawn without replacement, the same card cannot appear twice in one spread.' This is a key behavioral detail.
• Document the error response format in the tool definitions or server documentation. Specify that errors return {isError: true, content: [{type: 'text', text: '...'}]} so LLMs know how to parse failures.
• Add a 'scope: read:tarot' declaration to each tool (or similar) to enable permission-based agent configuration and audit trails.