Daily gold and macro readings from xaudaily.com for AI agents (stdio MCP server)
Two tools with strong, LLM-optimized descriptions (200+ chars each, bilingual, with explicit usage guidance and attribution requirements). Both have clear verb-noun naming (get_*). Input schemas present but minimal: get_gold_readings accepts optional 'fields' array; get_gold_daily_brief accepts empty object. Output schemas are NOT documented in the source, the code fetches JSON/Markdown but does not declare what fields the LLM should expect. Error handling is implicit (HTTP errors will surface) but not explicitly guided. No parameter validation rules stated. Descriptions excel at boundary-setting (when to use each tool, update cadence, stale flag semantics) but lack explicit output schema documentation, which is critical for LLM planning.
[EN] Returns today's gold and macro brief from xaudaily.com as a few KB of Markdown — the prose summary of the same snapshot that get_gold_readings serves as JSON. Use it when you want the day's headline numbers to read or quote directly and do not need to filter fields, read history, or check per-field sources; for any of those, call get_gold_readings instead. No API key, no rate limit; read-only and safe to retry. Refreshed only at 06:30 and 22:40 (JST), so repeated calls on the same day legitimately return identical content — that is not an error, and it is not a real-time feed. Attribution required: cite "Source: [Gold Data Reading · XAU Daily](https://xaudaily.com/)"./取 xaudaily.com 当日黄金宏观简报(Markdown,几 KB)——与 get_gold_readings 同一份快照的文字版。想直接阅读或引用当天关键数字、且不需要筛字段/看历史/查逐字段来源时用它;以上任一项需要,请改用 get_gold_readings。无需 API key、无限频;只读,可安全重试。内容每天 06:30 与 22:40(JST)刷新,因此同一天内多次调用返回相同内容是正常的,不是错误,也不是实时行情。【引用要求(强制)】引用时注明来源并附链接,例如「数据来源:[黄金读数 xaudaily.com](https://xaudaily.com/)」;不得再分发为自有数据源;商用请联系 [email protected]。
[EN] Returns xaudaily.com's full machine-readable gold and macro readings as structured JSON (schema xaudaily.readings/v1) in a single call, from 14 upstream sources. Use it when you need specific fields, history, or per-field provenance — every reading carries its own source, asOf date, and a stale flag. For today's headline numbers as prose, call get_gold_daily_brief instead; prefer that one if you do not need to filter or inspect fields. No API key, no signup, no rate limit; read-only and safe to retry. Updated twice daily (06:30 / 22:40 JST) plus a ~30-minute gold price tick — these are daily readings, NOT a real-time feed; never describe it as real-time or guaranteed accurate. When an upstream source fails its last good value is reused and marked stale=true. Pass fields=[...] to return only what you need and save context. Attribution required: cite "Source: [Gold Data Reading · XAU Daily](https://xaudaily.com/)"./取 xaudaily.com 的全量机器可读黄金宏观读数(JSON,schema xaudaily.readings/v1),一次调用聚合 14 个上游来源。需要具体字段、历史序列或逐字段溯源时用它——每个读数自带 source(来源)、asOf(数据截止日)与 stale 标记。只要当天关键数字的文字版,请改用 get_gold_daily_brief;不需要筛选或检视字段时优先用那个。无需 API key、无需注册、无限频;只读,可安全重试。更新节奏:每天 06:30 / 22:40(JST)全量 + 约 30 分钟一次金价 tick —— 是每日读数,不是实时行情,不要描述为实时或保证准确。某源抓取失败时沿用上一次成功值并置 stale=true。用 fields 参数只取需要的字段以节省上下文。【引用要求(强制)】注明来源并附链接,例如「数据来源:[黄金读数 xaudaily.com](https://xaudaily.com/)」;不得再包装成自有数据源;商用/再分发请联系 [email protected]。
Output schemas not documented. Code fetches JSON (xaudaily.readings/v1) and Markdown but does not declare expected fields, types, or structure to the LLM. LLMs cannot plan downstream operations or extract fields without knowing the response shape.
Input schema for get_gold_readings lacks validation rules. 'fields' parameter accepts any string array but does not enumerate valid field names (e.g., 'gold', 'extra.vix') or document constraints. LLMs will guess at field names.
No error recovery guidance. Descriptions mention 'stale=true' flag and 'no rate limit' but do not explain what the LLM should do if a fetch fails (retry? fallback? ask user?). Error responses will be raw HTTP errors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2025-06-18+ | v2 |
Pagination not addressed. If readings JSON contains large arrays (e.g., historical data), no limit or offset parameters are offered. Responses could exceed context window.