A private, searchable tarot experience archive: MCP server + mobile-first web UI
Server has 8 tools with basic schemas and descriptions, but significant gaps prevent higher scores. Most tools lack parameter descriptions (e.g., draw_spread has spread_id and question but no guidance on valid values). The write_reading tool is a mega-tool combining 9 distinct actions (create, add_interpretation, add_reflection, etc.) into one interface, violating single-responsibility principle. Output schemas are not documented. Error handling is absent from visible code. Tool descriptions are present but generic (e.g., 'Draw a tarot spread' lacks context on when to use vs list_spreads). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Naming is reasonable (verb_noun pattern) but composition is poor.
Get the history of a specific card across all readings
Draw a tarot spread with the specified spread ID and question
Return the approved reference meaning, separate from personal history.
Get a specific reading by ID
列出已保存但还没有任何解读的牌阵,按最新优先返回聊天所需摘要。
List all available tarot spreads with their positions and guidance
Search readings with optional filters for query, card name, orientation, spread ID, and date range
write_reading combines 9 distinct actions (create, add_interpretation, add_reflection, update, create_insight, link_insight_evidence, revise_insight, set_insight_maturity, archive_insight) into a single tool with a polymorphic 'action' parameter. This violates single-responsibility and forces LLMs to reason about which action to invoke rather than selecting a tool by name.
Parameter descriptions are missing or minimal. draw_spread accepts spread_id and question but provides no guidance on valid spread IDs, expected question format, or constraints. search_readings has 8 parameters with only 1-2 word descriptions (e.g., 'Filter by card name'). LLMs cannot infer valid values without explicit constraints.
Output schemas are not documented. Tools return complex nested objects (e.g., draw_spread returns cards array with position_id, card_id, card_name, orientation) but the response structure is not formally specified. LLMs cannot plan downstream tool calls without knowing what fields to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Write a reading with various actions: create, add_interpretation, add_reflection, update, create_insight, link_insight_evidence, revise_insight, set_insight_maturity, archive_insight
No tool annotations despite clear risk levels. write_reading is marked WRITE but has no destructiveHint. list_spreads and search_readings are READ_ONLY but lack readOnlyHint. Agents cannot distinguish safe from unsafe operations without annotations.
Error handling is absent from visible code. No recovery guidance, no categorization of retryable vs fatal errors, no actionable error messages. If write_reading fails due to invalid spread_id or missing required field, the LLM receives no guidance on what to do next.