Skip to main content

EchoPai (回声投研)

io.github.echopai/echopai-mcpVersion 1.0.1

China equity & quantitative research MCP server with 50 tools for reports, filings & sentiment.

Status

Responds1507 msAnswered our handshake without sign-in.
Last checked
Nov 16, 2026, 01:28 UTC
Latency
1507 ms
Remotes responding
1 of 1
Sign-in
No sign-in
Hosting
Hosted outside the EU (detected)
Registry data as of
Nov 16, 2026, 12:08 UTC

Data from the official MCP Registry, checked by run.zero. How we check

Connect

Streamable HTTP
https://mcp.huishengpai.com/mcpResponds1507 ms
No sign-in

Claude Code

claude mcp add --transport http echopai-mcp https://mcp.huishengpai.com/mcp

Claude Desktop and claude.ai

Settings > Connectors > Add custom connector
Name: echopai-mcp
Remote MCP server URL: https://mcp.huishengpai.com/mcp

Cursor

{
  "mcpServers": {
    "echopai-mcp": {
      "url": "https://mcp.huishengpai.com/mcp"
    }
  }
}

VS Code

code --add-mcp '{"name":"echopai-mcp","type":"http","url":"https://mcp.huishengpai.com/mcp"}'
Streamable HTTP
npm: @echopai/mcp 1.0.1
Package only

npm (npx)

claude mcp add echopai-mcp -- npx -y @echopai/mcp@1.0.1

Built your own server? run.zero hosts MCP servers on European infrastructure.

Host your own MCP server on run.zero

Tools

82 tools

Tools of EchoPai (回声投研)
Read-only
Idempotent
Open world
announcements_detailRead one announcement on demand (id from feed/stock/search `items[].id`) — an annual report is 200k+ chars, so do not expect the whole body. Financial question path: 1) `format=search` with `q` (space-separated, all terms must match) → hits with `section_path`, `scope_hint`, `table_index`; 2) `format=table` with that `table_index` → table + `unit` / `page` / `footnote`. Default `format=outline` gives headings (`char_offset`) and page anchors (short notices include full `content`); `format=full` pages via `offset` / `next_offset`. ⚠️ `scope_hint` is a hint: balances appear in both consolidated and parent-company notes, and `policy` hits are accounting policy text — say which you cite. `content_status`: `pending` = retry later, `unavailable` = read `pdf_url`. ⚠️ TIME: `published_at` is when the filing became visible; `disclosure_date` (next day 00:00) is never a publication time.Not declaredNot declaredNot declaredNot declared
announcements_feedRecent A-share announcement feed (sorted by published_at desc). Filter by `type` slug (e.g. annual_report / equity_change / restructuring) or `since` lower bound. Use when surveying market-wide disclosures over a recent window — for a specific stock prefer `announcements_stock`. ⚠️ TIME: `published_at` is when the filing actually became visible — most A-share announcements go live after the previous session's close, so a filing dated "2026-08-27" by the exchange typically became readable at ~18:00 on 2026-08-26. The scheduled-disclosure stamp is the *next* calendar day at 00:00 and is returned separately as `disclosure_date`; never read `disclosure_date` as a publication time, and never conclude a filing was published at midnight.Not declaredNot declaredNot declaredNot declared
announcements_searchHybrid search over A-share announcements: Chinese company name / 6-digit code / free-text keyword in one query (name or code → filter by stock; keyword → full-text relevance × recency over title + ai_summary + body start). Use when you DON'T have a canonical_code ("新华医疗 回购", "600587", "回购"); for a known stock prefer `announcements_stock`, for the market feed `announcements_feed`. ⚠️ TIME: `published_at` is when the filing became visible (usually the evening before the exchange date); `disclosure_date` (next day 00:00) is never a publication time.Not declaredNot declaredNot declaredNot declared
announcements_stockAnnouncement history for a single A-share security. Default lookback 30 days, max 5 years. AH dual-listing: pass the A-share canonical (announcements are A-share only — HK side has separate HKEX disclosures not covered here). ⚠️ TIME: `published_at` is when the filing actually became visible — most A-share announcements go live after the previous session's close, so a filing dated "2026-08-27" by the exchange typically became readable at ~18:00 on 2026-08-26. The scheduled-disclosure stamp is the *next* calendar day at 00:00 and is returned separately as `disclosure_date`; never read `disclosure_date` as a publication time, and never conclude a filing was published at midnight.Not declaredNot declaredNot declaredNot declared
auctionOpening call auction (集合竞价) as its own dataset — today or any past trading day. Items: `canonical_code`, `name`, `auction_price` (= the official open), `prev_close`, `pct_change` (points), `volume` (shares), `amount` (CNY). `stage`: `final` = settled result (after the 09:25 match; backfilled for history) / `live` = today 09:15–09:25 from the live snapshot — before 09:25 that is the INDICATIVE price, `volume`/`amount` null (nothing traded yet). Auction leaderboard: `sort=pct_change` + `limit` (`order=asc` for the weakest); watchlist: `codes`. Empty with a `note` on non-trading days or before the dataset starts. 竞价未匹配量 unmatched 为股,正买负卖;unmatched_as_of 是最后有效竞价观测的源时间(09:25 后冻结),历史未采集为 null。Not declaredNot declaredNot declaredNot declared
bars_batchBatch K-line for multiple A-shares in one round-trip. Daily bars default to date desc; use order=asc for chronological calculations. Response sort echoes the date order. Daily: ≤100 codes × ≤1 yr; minute: ≤20 codes × ≤7 d. Envelope: `{items:[{canonical_code, bars}], errors:[{canonical_code, error_code, message}]}`. ⚠️ Read `errors[]` first: an unknown code lands there as `unknown_security`, while a real security with no trading in the window returns an empty `bars[]` — report that as no data, not a bad code.Not declaredNot declaredNot declaredNot declared
calendar_reportsEarnings-report disclosure window for one A-share: when the next periodic report is booked (`next_report`), whether today is its booked disclosure date (`discloses_today`), and the most recent report actually filed (`latest_disclosed`). Per-period `status`: scheduled | due_today | awaiting_confirmation (booked date passed, actual date not confirmed yet — T+1 lag or deferral, NOT a missed filing) | disclosed | unscheduled. Always read `coverage` first: schedule_pending means the exchange has not published the next period's calendar yet (released only ~5 days before period end) — a normal blackout, NOT 'this company does not report'; unknown means the name is absent from the calendar. T+1 data.Not declaredNot declaredNot declaredNot declared
calendar_upcomingScan A-shares whose periodic report is booked for disclosure in the next N days, sorted by booked date — the overnight-risk screen for earnings season. Pass `codes` (comma-separated) to narrow to a portfolio, `report_type` to restrict the period. Only names not yet confirmed as disclosed are returned and the window starts today, so `awaiting_confirmation` names (booked date already passed) are deliberately excluded. Note each period's calendar is published only ~5 days before period end, so scanning well ahead of that returns few or no rows.Not declaredNot declaredNot declaredNot declared
capabilitiesPass `verb=<tool name>` to get that tool's RESPONSE FIELD LIST locally — no request, no credits; do this before writing a projection instead of guessing field names. Without `verb`: the API `spec_version` — NOT this CLI's version, never infer one from the other — plus a newest-first changelog of capability changes; call at session start or when a response shape surprises you. Free, unauthenticated.Not declaredNot declaredNot declaredNot declared
chartSingle-security K-line for **stocks and A-share indices** (daily by default; `minute=true` + one `date` for intraday). Multi-code: `bars_batch` (stocks only). Indices are detected from the code: `SSE:000xxx`, `SZSE:399xxx`, `BSE:899xxx`. ⚠️ The exchange prefix decides it — `SSE:000001` is the Shanghai Composite, `SZSE:000001` is 平安银行; pass `kind` to override. To measure a stock against its market, call twice over the same window (stock + index) — `index_snapshot` has no history. Index minute bars are capped at 7 calendar days. Indices add `pct_change` and have no turnover-rate or market-cap field.Not declaredNot declaredNot declaredNot declared
concept_tags查询股票当前热炒题材与完整基础概念标签,同时返回待确认、未同步、无已确认归属等状态。热炒活跃状态与长期基础归属分别返回;基础标签不等于当天上涨原因。Not declaredNot declaredNot declaredNot declared
concepts_alertsCurrently active concept alerts. Two rules: `big_move` (|pct_change| > 3%) and `limit_up_cluster` (limit_up_count >= 3 AND stock_count >= 5). Cheap, non-billable. Use as the agent's signal for 'which themes are running hot right now'. With `kind=industry`, only `big_move` ever fires — a Shenwan level-1 industry holds 150+ members, so the limit-up cluster rule would trigger nearly every day and drown the feed.Not declaredNot declaredNot declaredNot declared
concepts_alerts_historyConcept-alert historical events (default last 24h). Filter by `rule` (big_move / limit_up_cluster) and Unix-ms `from`/`to` window. Used by AI agent / admin to review what fired and when.Not declaredNot declaredNot declaredNot declared
concepts_daily_barsDaily OHLC for a concept index (chain-linked equal-weight, base 1000). Returns breadth fields and `is_backfilled` flag. Pass `kind=industry` when the id is a Shenwan industry — a dimension mismatch returns 404, same as an unknown id. Requires `bars:read` scope.Not declaredNot declaredNot declaredNot declared
concepts_fundflow_minuteMinute-by-minute capital flow for a concept or industry, summed over its current active constituents — the same basis as `main_net_in` in `concepts list`. Same field semantics as the per-security `fundflow` verb: `*_cum` are running day totals (never sum across bars), `*_delta` is the per-bar change. Amounts are CNY, positive = net inflow. Each bar carries `covered_members`, how many constituents had data by that minute. Pass `kind=industry` when the id is a Shenwan industry. Use it to see whether a theme is being accumulated or distributed while it moves.Not declaredNot declaredNot declaredNot declared
concepts_listList concepts with a live snapshot (pct_change / index_value / up_count / down_count / limit_up_count / amount / speed_3min). Default sort pct_change desc; `order=asc` gives the bottom-N instead (e.g. today's worst concepts). All sort keys (`pct_change` / `amount` / `main_net_in` / `limit_up` / `speed` / `stock_count`) rank the whole universe before `limit`, so the result is a true top-N (or bottom-N); concepts missing the sort key always go last. Entry point for browsing; drill in with `concepts_show` / `concepts_daily_bars`. Every item echoes `kind` (`concept` / `base` / `industry`) — pass it along with the id, since by-id verbs default to `kind=concept` and an industry id without it returns 404. During 09:00–09:14 values are the previous close (`note` says so).Not declaredNot declaredNot declaredNot declared
concepts_minute_barsConcept-index minute-level OHLC (chain-linked equal-weight, base 1000). Minute-end labels, 240 bars per full session (09:31-11:30, 13:01-15:00); boundary bars are merged preserving volume and amount. Max 7 days per request. Pass `kind=industry` when the id is a Shenwan industry — a dimension mismatch returns 404, same as an unknown id. Requires `bars:read` scope.Not declaredNot declaredNot declaredNot declared
concepts_newsRecent news mentioning the concept name (ILIKE on news.title + content). Short-term implementation; long-term will switch to AI tagging. Requires `news:read` scope.Not declaredNot declaredNot declaredNot declared
concepts_of**Which themes does this stock belong to?** Inverse of `concepts_show`: returns a security's active concepts (id, name, today's percent change). Pass `code` (preferred; `lookup` first if you only have a name) or `query` (substring over name/ticker, may match several — read each `canonical_code`); at least one. ⚠️ A concept with null `name` and `pct_change` exists but has no snapshot yet. ⚠️ `kind` returns one dimension: `concept` (hot-theme library, default), `base` (base labels), or `industry` — carry the dimension with each id. ⚠️ An empty array means no recorded membership in this dimension. Use `concept_tags` for both layers, coverage states and currently active themes.Not declaredNot declaredNot declaredNot declared
concepts_showDetail for one concept: meta plus current member securities with latest snapshots. Use after `concepts_list`. Pass `kind=industry` for a Shenwan industry id — a dimension mismatch returns 404, same as an unknown id. Members carry `close` / `pct_change` plus `price` (identical to `close`, the name `quote` / `scan` use for the last price). The response echoes `kind`; carry it with the id. Members are ordered by `main_net_in` desc by default (who is drawing money in); change with `member_sort` / `member_order`.Not declaredNot declaredNot declaredNot declared
concepts_snapshotBulk live snapshot of all concepts. Use `codes` (comma-separated concept_id list) to filter; omit for the full set. Cheaper than `concepts_list` when only the live numbers are needed without joining meta. Percent-change fields are in **percentage points** (`8.94` = +8.94%), same convention as `quote`.Not declaredNot declaredNot declaredNot declared
concepts_viewsBroker views associated with a concept (concept_views × broker_views, ai_status=done), sorted by published_at desc. Primary research source for a theme. Requires `views:read` scope.Not declaredNot declaredNot declaredNot declared
digestOne-shot research digest for one security: views (PRIMARY), quote, valuation snapshot, market sentiment, supplementary news, ownership_events (30-day insider trades + buybacks, next-30-day lockups) and margin (~7 trading days). Margin is **T+1** — judge direction by `financing_net_buy_cny`, never gross `financing_buy_cny`. Separate buckets, not ranked against each other; partial failures go to `meta.partial_failures[]`. First call for 'what's going on with <stock>'. A-share returns all buckets; HK/US return only views + news. A+H: pass the A-share canonical.Not declaredNot declaredNot declaredNot declared
dragon_tiger_dailyOfficial dragon-tiger list (龙虎榜) for one trading day, ranked by top-five seats' net buying: brokerage branches, institutional seats (机构专用) and northbound desks. **Published the evening after the close — during trading hours it returns the PREVIOUS trading day**; always state the response's `trade_date`. Check `coverage` per exchange (`complete` / `complete_empty` / `pending` / `failed`) before concluding a name is absent. A name can appear once per qualifying reason (`reason_category`) — do not dedupe. ⚠️ `net_buy_cny` nets the top-five DISCLOSED seats only; not comparable with moneyflow. ⚠️ Totals aggregate each side's OWN list, whereas vendors aggregate over the UNION of both lists, so vendor totals run larger — reconcile on `seats[]`, not on totals, and do not report the gap as a data error. ⚠️ Shanghai (incl. STAR) discloses one side per seat: `sell_cny = null` on a buy-side seat means NOT DISCLOSED, not zero. Prefer the precomputed `institution_net_cny` / `has_northbound`. Amounts CNY, ratios percentage points.Not declaredNot declaredNot declaredNot declared
dragon_tiger_seatOne seat deduplicated dragon-tiger appearance history: which names it traded, on which side, for how much, and how each opened the next session. Deduplication and side semantics match `dragon_tiger_seats` — one row per (trading day, security, side), amounts are the seat OWN side only, and `null` means the exchange did not disclose that side rather than zero. Use it to check whether a name on today board is a seat regular before reading anything into a single appearance.Not declaredNot declaredNot declaredNot declared
dragon_tiger_seatsRanks brokerage branch seats (营业部) by dragon-tiger activity: appearances, cumulative buying, and how their picks opened the NEXT session. Counting is deduplicated by (day, exchange, security, seat). ⚠️ Read `win_rate_pct` with `win_samples` — small samples are noise (`min_appearances` gates win-rate sorting). ⚠️ 机构专用 and 沪/深股通专用 are shared placeholder names, not individual seats — excluded by default; request them explicitly and check `is_aggregate`. No per-seat NET figures (Shanghai discloses one side per seat).Not declaredNot declaredNot declaredNot declared
dragon_tiger_stockOne security's dragon-tiger (龙虎榜) appearance history, newest first, each with its full buy/sell seat table — is the money institutional, northbound or retail-branch hot money, and do the same seats keep coming back. Caveats as `dragon_tiger_daily`: end-of-day data (the newest row is never today's flow); one row per qualifying reason; `net_buy_cny` covers the top-five disclosed seats only; totals aggregate each side's OWN list, smaller than vendor UNION totals — reconcile on `seats[]`; Shanghai `null` = not disclosed, not zero. Empty = not on the list in the visible window (depth follows plan tier). Amounts CNY, ratios percentage points.Not declaredNot declaredNot declaredNot declared
earnings_recentScan A-share earnings pre-announcements filed in the last N days, newest first then YoY floor desc — the first-hand material for attributing a move to results rather than news flow. Filter by `direction` (positive|negative|uncertain), `report_type`, `min_pct` (YoY floor in percentage points), or `codes` to narrow to a portfolio. `revision` is computed against each name's FULL history, not just rows inside the window, so a downgrade filed today reads as revised_down even when its first version predates the window. Note `min_pct` also drops rows that have no percentage data.Not declaredNot declaredNot declaredNot declared
earnings_showEarnings pre-announcements and flash reports for one A-share. `forecasts` = mandatory guided ranges (net-profit floor/ceiling CNY + YoY); `express` = voluntary unaudited actuals. `direction` (positive / negative / uncertain) comes from the forecast TYPE, not the percentage (turnaround names have misleading YoY). Every revision is kept with `revision` (first / revised_up / revised_down / revised_flat); `has_downward_revision` flags a downgrade. ⚠️ Flash reports are voluntary and sparse — an absent `express` is normal.Not declaredNot declaredNot declaredNot declared
etf_flowsRanks ETFs by estimated net creation/redemption in CNY over a 1/5/20 trading-day window. **An estimate** (`share change × that day's unit NAV`), not a reported flow, and **null is not 0**: uncomputable days (stale shares / NAV, share-conversion dates) are skipped — **read `valid_days` with the total**. ⚠️ Without `asset_class`, money-market ETFs dominate by one to two orders of magnitude; `asset_class=equity` fixes that but drops every Shanghai-listed fund.Not declaredNot declaredNot declaredNot declared
etf_listDirectory of A-share ETFs: manager, tracked index, tick size, daily limit band. Filter by `index_code` (`000300` or `SSE:000300`), `q` (name/ticker substring) or `asset_class`. For a fund's own numbers use `etf_show`. ⚠️ `asset_class` is only published for Shenzhen-listed funds — filtering by it drops every Shanghai one. ⚠️ `limit_pct` null = band unknown, not unbanded. `price_tick` is 0.001 (a tenth of the stock tick).Not declaredNot declaredNot declaredNot declared
etf_showOne ETF's profile plus its latest daily row: close, shares outstanding, unit NAV, premium/discount, estimated net creation. ⚠️ Two premium fields, not interchangeable: `premium_rate_same_day` only when the NAV date equals the trade date; `premium_rate_ref` falls back to the latest NAV (date in `premium_ref_nav_date`) — cross-border ETFs publish NAV late, so `ref` is not same-day. Both are ratios (0.012 = 1.2%), not percentage points. ⚠️ `accum_nav` is not on the same basis as `unit_nav` after a share conversion — do not compare or derive returns. `quality` / `inputs_mask` say which inputs were present; `latest` null = no row yet.Not declaredNot declaredNot declaredNot declared
events_timelineKEY FIELDS (per row in `upcoming[]` / `past[]`): `event_kind` (not `kind`), `title`, `expected_on` + `date_precision` + `window_end`, `entities[].entity_code`, `importance`, `status`. One A-share security's dated events as two lists: `upcoming` (nearest first) and `past` (most recent first, within the look-back). `limit` applies to each list separately. ⚠️ Only `date_precision = day` rows are pinned to a day; for week/month/quarter/half/year rows `expected_on` is the START of the window (`window_end` is the far edge) — never render it as the event date. ⚠️ `expired` does NOT mean it did not happen — the window closed without confirmation reaching us; `cancelled` means withdrawn. `certainty: confirmed` on a research-sourced row is not an exchange filing.Not declaredNot declaredNot declaredNot declared
events_upcomingKEY FIELDS (per item): `event_kind` (not `kind`), `title`, `summary`, `expected_on` + `date_precision` + `window_end`, `entities[].entity_code` (the canonical code lives here), `entities[].role`, `importance` (1–5), `status`, `certainty`, `scope`. Forward catalyst calendar for the whole A-share market: earnings disclosures, lock-up expiries, buyback deadlines, dividends, IPOs, LPR, macro / policy calendar, major tech conferences, US earnings for A-share supply chains, and forward milestones from broker research and newswires. Answers what is scheduled, never what happened. Filter with `kinds` (prefix wildcards like `macro.*`), `scope`, `codes`, `importance_min`, `statuses`. Pass `precision=day` unless you handle `window_end` yourself. ⚠️ Only `date_precision = day` rows are pinned to a day; for week/month/quarter/half/year rows `expected_on` is the START of the window (`window_end` is the far edge) — never render it as the event date. ⚠️ `expired` does NOT mean it did not happen — the window closed without confirmation reaching us; `cancelled` means withdrawn. `certainty: confirmed` on a research-sourced row is not an exchange filing.Not declaredNot declaredNot declaredNot declared
financials_pitPoint-in-time financial indicators for one A-share at a given trade_date. Returns the latest report visible AS OF that date (announce_date ≤ date; conservative 90-day fallback when announce_date is missing). Designed for backtests / AI agents to avoid look-ahead bias. ~25 fields incl. EPS / BPS / ROE / margins / revenue / net-income / equity.Not declaredNot declaredNot declaredNot declared
financials_quote_snapshotOne-call 14-field valuation snapshot for one A-share: PE / PE-TTM / PB / PS / PS-TTM, total/float/free share, total/circ market cap, turnover rate (raw & float), volume ratio, dividend yield (last-year & TTM). Computed in-house from raw market data rather than passed through from a vendor feed, and cross-checked against an independent reference. Use `date` to fetch a historical snapshot; omit for real-time.Not declaredNot declaredNot declaredNot declared
financials_reportsLast N report-period snapshots for one A-share (~25 fields per period). Each item carries `announce_date` for visibility timing. Filter by `kind` to scope to Q1 / H1 / Q3 / annual / preliminary.Not declaredNot declaredNot declaredNot declared
financials_seriesTime series of a single financial metric for one A-share across reporting periods. `metric` is an indicators-table field name (roe_simple / revenue / ni_parent / debt_asset_ratio / gross_margin / eps_basic, ~150 supported).Not declaredNot declaredNot declaredNot declared
fundflowMinute-by-minute capital flow for one A-share, split by order size: main force (super + big), super, big, normal, small, retail. Amounts are CNY; positive = net inflow. `*_cum` fields are running totals for the day — never sum them across bars; `*_delta` is the per-bar change and is what you sum. Shanghai and Shenzhen only, from 2026-05-26; intraday it runs to the latest completed minute. Use it to tell whether a move is carried by large orders or by retail, and to spot late-session accumulation a daily number hides.Not declaredNot declaredNot declaredNot declared
global_index_snapshotSnapshot for 30 global equity indices plus Hong Kong indices (e.g. 日经225 GIDX:NKY, 标普500 GIDX:SPX, 纳斯达克 GIDX:IXIC, 恒生指数 HKEX:HSI). A-share indices are `index_snapshot`. These markets span time zones — always read `data_tier`: `realtime` = trading now; `closed` / `warm` / `daily` / `stale_cache` = a previous close, never report as live; `unavailable` = no data. `is_stale` is true for everything except `realtime`; `meta.realtime_count` / `meta.closed_count` give the split. `market_trade_date` is each market's LOCAL trading day — never compare it with a Beijing date. `volume` null = not published.Not declaredNot declaredNot declaredNot declared
hotSHAPE: `data` is a TOP-LEVEL ARRAY of `{name, canonical_code, mention_count, latest_at}` — no `items` wrapper (write `[*].name`). Stocks most mentioned by sell-side research notes; array order is the rank. `mention_count` is every mention in the window, not deduplicated. ⚠️ Sell-side attention, not retail — retail attention is `social_hot`; the two do not validate each other. HK names appear and can lead (`HK:` prefix). Window: trailing `hours` (default 24), or with `date` the previous trading day's 15:00 → that day's 15:00 (China time). ⚠️ Walk trading days, not calendar days — a weekend `date` returns a cumulative, overlapping window; drive the walk from `trading_days`. ⚠️ Windows differ in length (a Monday spans the weekend), so `mention_count` does not compare across days as-is; rank within a day does. An empty array means no mentions (failures return 503). History starts 2026-02-25 — earlier dates are *not covered*, not quiet.Not declaredNot declaredNot declaredNot declared
index_snapshotReal-time snapshot for the 172 A-share indices with a live quote feed (e.g. 上证指数 SSE:000001, 深证成指 SZSE:399001, 创业板指 SZSE:399006, 科创50 SSE:000688, 北证50 BSE:899050): price, pct_change, amount, speed_3min per index, refreshed ~15s during market hours (last intraday snapshot outside trading hours). Pass `codes` to narrow to a subset; omit for all 172. Use this for 'how is the broad market doing right now' — `quote` covers individual stocks only. CSI-series indices (CSI:*) have no live feed and are not returned.Not declaredNot declaredNot declaredNot declared
limit_up_analysisPer-stock limit-up attribution based only on published views and research. Returns leading_concept, concept_tags, nullable reason, attribution_status, empty_reason, evidence_refs and last_attempt_at. Incremental updates depend on service configuration; scheduled generation also maintains processing status. pending, processing, no_verified_catalyst, failed, budget_deferred and legacy_unknown are distinct. Inherited reasons retain their original generation time. Omit trade_date for the latest stored day; pass today explicitly for today's coverage. Optional as_of reads saved versions available at a historical instant without using later evidence.Not declaredNot declaredNot declaredNot declared
limit_up_historyDaily limit-up trend over the last N trading days: { trade_date, limit_up_count, broken_count, ever_broken_count, max_height }, oldest first. Same two seal-break measures as `limit_up_summary` (instantaneous end-of-day vs cumulative ≥30s breaks) — different populations, NOT nested; never subtract the series. ⚠️ The counting rule changed twice: values before mid-June 2026 include auction noise and days before 2026-08-25 count brief flicker — expect a STEP DOWN at 2026-08-25; segments are not comparable.Not declaredNot declaredNot declaredNot declared
limit_up_poolA-share limit-up pool for a trade date, per stock: first/last seal time, final pct, seal amount (CNY) / volume (lots), `max_seal_amount` + `max_seal_time` (the day's peak seal), `broken_count`, `consecutive_days` (连板数), `board_count_within`, one-word-board flag, status (limit_up / broken), board class. Read `max_seal_amount / seal_amount`: peak >> current = 封单被吃掉, ≈ 1 = 封单从来就薄. ⚠️ The peak has no history before 2026-08-25 (0 / null = not recorded — never substitute `seal_amount`). `broken_count` = 当日开板次数, debounced: counts breaks below the limit lasting ≥30 consecutive seconds after a genuine intraday seal (auction touches never count); kept after re-sealing. ⚠️ Our own measure — not comparable with terminals' 开板次数, not with the `broken_count` of `limit_up_summary`, and not across 2026-08-25 (older dates used a per-snapshot rule). 行内提供 pct_change(兼容 final_pct)、turnover_rate、amount、main_net_in、super_net_in,资金流为当日最新已存分钟累计元;metrics_trade_date、quote_as_of、fundflow_as_of 标明日期和源时间,缺测为 null,历史不混入今日数据。 Rows include same-day `reason` / `concept_tags` / `attributed_at`, preferring LLM `leading_concept` while keeping `graph_leading_concept`; attribution outages are explicit. `prev_trade_date` and nullable `resumed_after_suspension` distinguish recorded suspensions from data gaps. Sort: consecutive_days → board_count_within → seal_amount, desc. Pre-open fallback applies.Not declaredNot declaredNot declaredNot declared
limit_up_promotion晋级 / 断板追踪: every stock that closed sealed at the limit on the previous trading day, and what it did on `trade_date` — grouped by yesterday's height (highest first). Each level: `count` / `promoted` (sealed again, `today_height` = height+1) / `broken` (touched but unsealed) / `failed` (never reached the limit) / `promotion_rate` / `avg_pct_change` (that cohort's premium today). `members[]` carry `canonical_code`, `name`, `prev_height`, `outcome`, `today_height`, `pct_change`. The core sentiment-cycle read (昨日 3 板今天几家晋级、几家炸、几家跌停) that `max_height` cannot give. Intraday the outcome is live and can still change before the close.Not declaredNot declaredNot declaredNot declared
limit_up_summaryLimit-up summary for a trade date: limit_up_count / broken_count + ever_broken_count / limit_down_count / max_height / ladder[{height, count}]. Two seal-break measures: `broken_count` = INSTANTANEOUS boards open right now (day high touched the limit, currently unsealed; auction touches included); `ever_broken_count` = CUMULATIVE 当日曾炸板家数 (sealed ≥9:30 then broke for ≥30 consecutive seconds, re-sealed included). ⚠️ Different populations, NOT nested — ever_broken_count can be lower than broken_count; never subtract them. '现在多少没封住' → broken_count; '今天封过又开过' → ever_broken_count. ⚠️ Our own measure, not comparable with terminals, and not across 2026-08-25 (rule change). Pre-open fallback applies.Not declaredNot declaredNot declaredNot declared
lookupSHAPE: `{matches: [{canonical_code, name_cn, ticker, exchange}], count}` — the row container is `matches`, not `items`, and the name field is `name_cn`. Resolve a Chinese name / A-share code / pinyin initials to canonical codes (e.g. SSE:600519). Use this first whenever the agent has a description but needs a canonical_code for downstream calls. AH dual-listing: when a company is listed on both A-share and HK (e.g. 工商银行 SSE:601398 / HK:01398), prefer the A-share canonical for all downstream analysis unless the user explicitly asks for the HK side.Not declaredNot declaredNot declaredNot declared
margin_collateralDaily collateral-eligible holdings (担保券) series for one A-share: `collateral_shares`, `collateral_shares_chg`, `prev_collateral_shares`, `collateral_mv_cny`, `total_mv_cny`, `collateral_pct_of_total_mv` (collateral value as % of total cap, percentage points — a high and rising share means more of the float is exposed to forced liquidation). **T+1, published later than `margin_show`** (may lag by a day). Coverage is wider than the margin list, so presence here without `margin_show` is not a gap. `null` = not published, not zero.Not declaredNot declaredNot declaredNot declared
margin_collateral_moversRanks A-shares by collateral-eligible holdings (担保券) for one trading day (default latest). Default `pct_desc` = `collateral_pct_of_total_mv` (collateral value as % of total cap, percentage points) — how much of a name sits in margin accounts and is exposed to forced liquidation. Other keys: `pct_asc`, `shares_chg_desc` / `shares_chg_asc`, `mv_desc`. Rows with an unpublished ranking field are excluded, not zero. **T+1, published later than margin data** — may lag `margin_movers` by a day. ⚠️ Noisy on small caps — check `collateral_mv_cny`. Coverage is wider than the margin list.Not declaredNot declaredNot declaredNot declared
margin_marketMarket-wide margin-trading (融资融券) series, two arrays. `exchanges`: per-exchange daily summary (SSE / SZSE / BSE) — balances, gross buy, repayment, net buy, short-sell, `derived_fields`. `structure`: market-wide leverage metrics, most importantly `avg_maintenance_ratio` (margin-call pressure; below 1.3 is the forced-liquidation line), plus balance vs free-float cap, short-sell value, margin share of turnover, account counts and collateral value. **T+1**, and `structure` publishes later — the two arrays may not end on the same date; read each row's `trade_date`. Direction = `exchanges[].financing_net_buy_cny`; `structure[].financing_buy_cny` is gross.Not declaredNot declaredNot declaredNot declared
margin_moversRanks A-shares by margin-trading activity for one trading day (default latest). Sort: `net_buy_desc` (default) / `net_buy_asc` / `balance_desc` / `balance_chg_pct_desc` / `balance_chg_pct_asc`. Ranking by gross buy is deliberately not offered (it would surface net-outflow names as leveraged buying). **T+1** — the ranked day is normally the previous trading day. `balance_chg_pct_*` is noisy on small balances — check `financing_balance_cny`.Not declaredNot declaredNot declaredNot declared
margin_showDaily margin-trading (融资融券) series for one A-share: financing balance / gross buy / repayment / net buy, short-sell and short balance (shares and CNY), total balance, day-over-day change. **T+1 — the latest row is normally the previous trading day** (in backtests visible_date = trade_date + 1). ⚠️ `financing_buy_cny` is GROSS (includes round-trips) — judge direction by `financing_net_buy_cny` only; short side likewise by the change in `short_balance_shares`. ⚠️ `null` = not published / not derivable (Shenzhen and Beijing publish no repayment), not zero. ⚠️ `financing_balance_chg_*` is versus `prev_trade_date` — check it is adjacent.Not declaredNot declaredNot declaredNot declared
market_calendar按显式 market 查询交易日历,当前仅接入 HK,CN/US 预留且不回退。含完整交易日、半日市及休市日。最多 366 天,days 为含首尾日期的列表。calendar_known=false 或 is_open=null 表示未覆盖,不能按工作日推测。is_settlement_day=null 表示交收属性未知。source 返回发布版本与最后成功检查时间。Not declaredNot declaredNot declaredNot declared
market_clock按显式 market 查询市场阶段,当前仅接入 HK,CN/US 预留且不回退。含半日市、竞价、午休及人工例外;省略 at 使用服务端当前时间。at 必须带时区,返回所选市场的 IANA 时区时间。trading_profile=index 只表示相关市场阶段,指数计算时段未核验。午间扩展交易须提供有已核验资格的证券代码。actual_end_unknown=true 表示随机结束时点未知,null 表示日历或阶段未知;不存在的具体证券返回 security_not_found;市场阶段不证明个股未停牌。Not declaredNot declaredNot declaredNot declared
market_digestOne call for the whole board: session, 6 core indices, breadth + turnover, limit-up premium & ladder (with promotion rate), top concepts / industries and top 10 stocks by `pct_change` / `main_net_in`. Every list is `{items}`, fixed length — use the single verbs for longer lists. `mode` + `trade_date` say which day: before 09:15 or on a non-trading day it is the last session's close; 09:15–09:29 `top_pct_change` ranks the auction. Empty lists carry `empty_reason`; failed blocks are null + `meta.partial_failures`.Not declaredNot declaredNot declaredNot declared
market_indices按显式 market 查询指数目录,当前仅接入 HK,CN/US 预留且不回退。港股包含:恒生、国企、恒生科技、红筹、综合指数。items 行标识为 canonical_code;registry_revision 为随代码维护的注册表版本,不能视为实时行情源日期。quote_enabled/chart_enabled=false 表示报价和图表未启用。Not declaredNot declaredNot declaredNot declared
market_moversA-share top-N movers from the live snapshot (or a past trading day via `date`). Sort keys: `pct` / `pct_asc` / `speed` / `amount` / `turnover` / `total_mv` / `main_net_in` / `main_net_in_asc`; rows with a null sort value land last in both directions. Downloads the full ~3 MB snapshot and ranks client-side — `scan` with `sort` + `limit` ranks server-side and is cheaper. Empty with a `note` during 09:00-09:14. Each row carries `industry` and `top_concept`; both are often missing — coverage, not failure (`top_concept` a subset of A-shares; `industry` excludes `BSE:*`), and `top_concept` is not the limit-up `leading_concept`, which is not served here. Live rows also expose `top_base_concept` `{id,name,kind:base,pct_change}` and `top_concept_status` / `top_base_concept_status` (ready, no_membership, no_snapshot, unavailable). The two layers rank independently; use kind=base for base detail. Historical rows carry neither current theme field nor their statuses. The snapshot includes ETFs: pass `security_type=stock` for a stock-only `amount` board. Historical `date` boards sort only by `pct` / `pct_asc` / `amount` / `turnover`, carry no `top_concept`, and cannot combine `concept_id`.Not declaredNot declaredNot declaredNot declared
market_statusCurrent A-share session state. `session` ∈ `pre_open` (09:00-09:14) / `auction` (09:15-09:29) / `open` / `break` (lunch) / `closed`, same enum as `quote`. **Read `session`, not `is_market_open` / `is_premarket`** — both are false at lunch AND after the close. `trade_date` is today's China calendar date, not the trading day — qualify with `is_trading_day`; `next_trading_day` null = calendar not populated (unknown). `is_holiday` marks shut weekdays only (weekends read false). Free — call before quote / bars to choose pre-market vs intraday vs after-close behaviour.Not declaredNot declaredNot declaredNot declared
newsSUPPLEMENTARY news / market briefs (short-horizon event stream) — use only to fill gaps not covered by `views`, the primary research source. Default `fields=compact` (title + ai_summary + ~300 chars); `full` only when needed. Flash news has `title: null` — read `display_title` (falls back to the first sentence), or pass `has_title=true` to drop untitled items. Modes: `code` (canonical, A-share / HK / US) → news mentioning that security; `query` (free text, all terms required, newest first); neither → recent feed. A canonical_code passed as `query` will not match — use `code`. A+H: pass the A-share canonical.Not declaredNot declaredNot declaredNot declared
news_getFetch one news item by id — the read half of list→read, for when a `news` list row is not enough. Returns that item's full content plus its tagged securities and source labels. ⚠️ Field `author` is the **byline entity — an outlet brand or a social account handle, never a reporter's name** (reporter bylines are split off at ingest), so do not present it as a person; `source_label` is the collector-level source. Remember `news` is the SUPPLEMENTARY stream: for forming an investment opinion prefer `views`. Free, and takes no scope of its own.Not declaredNot declaredNot declaredNot declared
ownership_lockupsScan upcoming share-lockup expirations across the whole A-share market for the next N days, sorted by unlock date then unlock pct desc. Forward-looking sell-pressure radar that digest can't provide (digest is single-code). Filter small unlocks with min_pct (percentage points of total shares); pass `codes` to restrict to a portfolio instead of the whole market.Not declaredNot declaredNot declaredNot declared
ownership_showOwnership events for one A-share: recent insider holder trades (net shares/pct, increase/decrease counts), share repurchases (ongoing count, executed amount/shares, latest status) and upcoming lockup expirations (unlocking shares/pct, next date). Holder-trade / repurchase windows look back N days; lockups look forward. ⚠️ A repurchase price ceiling BELOW the current quote is almost certainly a restricted-share cancellation at grant price, not open-market support. ⚠️ `plan_expires_on` is usually null = UNKNOWN, not still valid — use `disclosed_days_ago`. T+1.Not declaredNot declaredNot declaredNot declared
quoteReal-time quote for 1-200 A-share securities (last price, volume, change %, bid/ask). For >200 codes use `scan`. HK codes are not supported; for an A+H company use the A-share canonical (SSE:601398, not HK:01398). Rows also carry `turnover_rate` (percent), `same_minute_volume_ratio` (cumulative shares through the snapshot minute divided by the five-session same-minute cumulative-share average; null if missing; not comparable to market turnover `volume_ratio`), `fundflow_latest` {trade_date, bar_time, main_net_cum, super_net_cum} (latest stored minute cumulative CNY; check its date/time), `industry` (Shenwan L1/L2), `top_concept` (strongest member concept now) and, for margin-list names, a T+1 `margin` block. Expect `industry` / `top_concept` to be absent on many rows — coverage, not failure: `top_concept` resolves for only a subset of A-shares and `industry` does not cover 北交所 (`BSE:*`). `top_concept` is **not** the limit-up LLM `leading_concept`, which is not served here. Live rows also expose `top_base_concept` `{id,name,kind:base,pct_change}` and `top_concept_status` / `top_base_concept_status` (ready, no_membership, no_snapshot, unavailable). The two layers rank independently; use kind=base for base detail. Historical rows carry neither current theme field nor their statuses. **In `pre_open` / `auction` sessions read the `auction` block `{status, prev_close, price, pct_change, volume, amount}`, not top-level `price` / `pct_change`.** `status` is `not_started` / `matched` / `no_match`; a top-level `pct_change` of 0 in that window is ambiguous — only `auction.status = matched` with `auction.pct_change = 0` is a flat auction. 盘前顶层 prev_close 与 auction.prev_close 对齐为今日基准,prev_close_date 为上一交易日;previous_close 为昨日实际收盘价。prev_close_basis 区分交易所参考价、事件调整价与未校准昨收,prev_close_provisional=true 时不可视为已确认除权参考价。未撮合的顶层 price 仅为基准展示,pct_change 为 null。 Suspended / frozen names report `no_match`; `auction.volume` / `amount` stay null until the 09:25 match (volume in shares). Read `markNot declaredNot declaredNot declaredNot declared
reportFetch the FULL research-report text behind a view (parsed long-form), for when the `views` summary is not enough. Only works on views where `full_text_available=true` (archived sell-side reports) — get those from `views --report-only`, do not guess a view_id. A 404 `no_full_text` means that one view is a plain opinion with no archived report; it does NOT mean the endpoint is unavailable or that no reports exist. Consumer pricing: 1 credit per PDF page, with no minimum or maximum document charge, per UTC day. Call format=quote first for free page_count and credit_cost; all subsequent outline/full reads share the same daily document charge. Unknown PDF page count blocks paid reads; unpaginated non-PDF documents cost 1. Two-step usage: default `format=outline` returns a few-KB heading map + total_chars + each heading's char_offset; then call `format=full` with `offset` (e.g. an outline char_offset) to read the body in pages — follow `next_offset` to page through large docs instead of pulling everything at once.Not declaredNot declaredNot declaredNot declared
scanFull-market real-time quote scan (~5800 A-share securities, ~3 MB unfiltered). Narrow server-side first: `sort` + `order` + `limit` return a real ranking (`sort=pct_change, order=desc, limit=50` = the day's 50 strongest in a few KB); `limit` without `sort` is an alphabetical sample, not a ranking. When rows are dropped the response carries `truncated: true` and `matched_total` (the pre-limit count) — `total` is only how many rows came back. The universe includes ETFs (`security_type` on every row); money-market ETFs dominate `sort=amount`, so pass `security_type=stock` for a stock-only board (filtered before sort/limit). `include_margin` attaches the T+1 margin block; `include_industry` attaches `industry` (Shenwan L1/L2) and `top_concept`. Both are often absent — coverage, not failure: `top_concept` resolves for only a subset of A-shares and `industry` does not cover `BSE:*`; `top_concept` is not the limit-up LLM `leading_concept`, which is not served here. Live rows also expose `top_base_concept` `{id,name,kind:base,pct_change}` and `top_concept_status` / `top_base_concept_status` (ready, no_membership, no_snapshot, unavailable). The two layers rank independently; use kind=base for base detail. Historical rows carry neither current theme field nor their statuses. **In `pre_open` / `auction` sessions read each row's `auction` block, not top-level `price` / `pct_change`** (same rules as `quote`: only `auction.status = matched` with `auction.pct_change = 0` is a flat auction; compute from `auction.prev_close`). Read `market_trade_date`, not `snapshot_at`, to decide whether a row is today's — a frozen feed keeps `snapshot_at` moving. `is_margin_target` is margin-list membership (T+1), not activity; the key is absent (never `false`) when the list cannot be loaded. `trade_date=YYYY-MM-DD` scans that past day's daily bars (see the parameter for allowed combinations). 竞价未匹配量 unmatched 为股,正买负卖;unmatched_as_of 是最后有效竞价观测的源时间(09:25 后冻结),历史未采集为 null。Not declaredNot declaredNot declaredNot declared
searchStructured-first hybrid search across analyst views + news (views weighted higher). Three lanes — entity (company codes / names, analyst names), concept-graph hub (theme alias → concept → constituents & research), and full-text keyword matching (Chinese word-segmented) — fused with RRF. Deterministic & explainable: no vector kNN, no LLM rerank. Best for theme/concept queries: 'CPO' brings out 光模块/光通信/光芯片; '减肥药' brings out 创新药; '智驾' brings out 智能驾驶. Use `news` or `views` instead if you need strict time-window listing.Not declaredNot declaredNot declaredNot declared
security_reference按规范代码查询证券参考详情,当前仅接入港股,CN/US 预留:ISIN、币种、每手股数、双柜台关联及身份核验状态。source 包含来源、版本、源日期、观测时间和最后成功检查时间;stale=null 表示未知。collection_eligible 仅为参考资格,不代表报价已启用。listing_status/trading_status=unknown 不可解释为正常交易。Not declaredNot declaredNot declaredNot declared
sentiment_breadthIntraday breadth time series for one trading day (240 minute bars; 13:00 excluded — ticks at the lunch-break boundary are too sparse). `trade_date` omitted = today (pre-open returns empty); explicit YYYY-MM-DD = historical replay. Returns per-minute rows of up/down/flat/limit_up/limit_down/breadth/divergence fields — full mode matches `sentiment_overview`; compact omits long summaries and extended breadth fields. Coverage reports source and returned time ranges. Budget trimming retains latest points, ascending.Not declaredNot declaredNot declaredNot declared
sentiment_overviewAggregate sentiment for one trading day's latest minute: up/down/flat, limit-up/down, gt3/lt3 breadth, divergence index + label, MA20/50/200 breadth, 52w new highs/lows. Omit `trade_date` for today realtime. Also a `leverage` block (融资融券): judge by `exchanges.financing_net_buy_cny`, never gross `financing_buy_cny`; `structure.avg_maintenance_ratio` below 1.3 = forced-liquidation line. **The leverage block is T+1 — not the same day as the rest** — and each sub-block has its own `as_of`. `leverage` null = unavailable. Also a `turnover` block (current_turnover / predicted_total / delta_vs_yesterday, same numbers as `sentiment_turnover`'s header) — no second call needed for 成交额. `turnover.volume_trend` = `expanding` (放量) / `contracting` (缩量) / `flat` vs the previous two sessions at the same minute (`vs_prev_2d_ratio`, per-day `prev_sessions[]`); read it instead of comparing against yesterday's full-day total.Not declaredNot declaredNot declaredNot declared
sentiment_pct_distributionReal-time A-share market pct-change distribution: bucket counts at ≤-10% / -9% / ... / +10% / 涨停, plus universe total. Snapshot only — historical replay not yet exposed.Not declaredNot declaredNot declaredNot declared
sentiment_premiumSHAPE: two curves `main[]` and `compare[]` (not `bars` / `items`), rows `{bar_time, avg_pct, stock_count}`, plus `main_pool_size` / `compare_pool_size`. Are yesterday's limit-up buyers making money today? Yesterday's limit-up pool (minus one-word boards and boards that broke and stayed broken), plotted minute by minute today as an **equal-weighted** average percent change; `compare` is the same curve one day earlier. ⚠️ Read `main_pool_size` / `compare_pool_size` with the curves — they tell a premium move from a population change. ⚠️ Equal-weighted, not cap-weighted, so not comparable to index returns. Series use 240 minute end labels (09:31-11:30, 13:01-15:00). Pool dates are trading days, resolved server-side. Empty `main` with a non-zero pool size = session not started. Optional scope filters both pools; omit it to preserve the legacy all_a basket. ETF is excluded.Not declaredNot declaredNot declaredNot declared
sentiment_turnoverIntraday turnover time series with `current_turnover` / `predicted_total` / `delta_vs_yesterday` headline fields, plus `volume_trend` (`expanding` 放量 / `contracting` 缩量 / `flat`) and `vs_prev_2d_ratio` vs the previous two sessions at the same minute. `trade_date` omitted = today (pre-open returns nulls); explicit YYYY-MM-DD = historical. `days` (1-5) pulls that day + N-1 prior trading days for same-minute YoY comparison.Not declaredNot declaredNot declaredNot declared
social_hotRetail-attention leaderboard: A-shares retail investors are watching now, aggregated across several public attention boards. ⚠️ Retail board — `hot` is the sell-side board; they measure different populations and do not validate each other. `score` / `mention_count` are a weighted rank aggregate, **not a count of posts** and not comparable to `hot`. Array order is the rank. `rank_change` is versus one hour ago (null = not on the board then). ⚠️ Snapshot only, no history. Read `computed_at` / `stale_seconds` for freshness. Attention is not a bullish signal.Not declaredNot declaredNot declaredNot declared
technicalsKEY FIELDS (per item): `pos_in_52w.value` (0–1, plus `high_52w` / `low_52w` / `breakout_52w` / `breakdown_52w`), `consecutive_up.days` + `.pct_change`, `volume_spike.median_next_ret_pct` / `.avg_next_ret_pct` / `.winrate` / `.sample_size` (no `next_day_median_pct`), `overhead.supply_days` / `.traded_amount_ratio`. Every block has `observed_days` + `status` (`no_data` = insufficient history). End-of-day technicals for up to 50 A-shares on one server-side methodology. Use scan / limit_up_pool to find candidates, this to vet them. Read winrate only with sample_size, the Wilson interval and excess_winrate — samples are often single-digit.Not declaredNot declaredNot declaredNot declared
trades_minute**Who is actually buying — from the trades themselves.** Per-minute (or 5/15/30/60-minute) aggregates over the raw trade feed for one A-share: buyer- vs seller-initiated volume, large-print counts, true VWAP. The self-verifiable counterpart to `fundflow` (the vendor's own classification) — the two routinely disagree. `buy_ratio` = buy_vol / (buy_vol + sell_vol); ⚠️ Neutral volume is deliberately excluded from the denominator. `vwap` is yuan per share. ⚠️ `big_count` / `big_amount` use a fixed 200,000-yuan-per-print line, meaningless on high-priced names (贵州茅台: 97 of 99 prints in one 5-minute bucket cleared it) — compare a stock with itself only; unrelated to fundflow tiers. ⚠️ `tick_count` counts the prints **we captured** (3-second snapshots), not the exchange's trade count. Omit `date` for the latest day that has data. History starts 2026-05-26; earlier = not covered.Not declaredNot declaredNot declaredNot declared
trades_ticks⚠️ **EXPERIMENTAL — may change or be withdrawn at any time.** Do not build automation on it or persist it as a system of record. Individual prints for one A-share on one trading day, in `seq` order with cursor paging — for inspecting a specific moment; for whole-session work use `trades_minute`. ⚠️ Not exchange tick data: one row is everything traded in a 3-second window, so to find large trades filter on `min_amount` — counting rows will not do it. ⚠️ `side` is inferred from bid/ask (`M` = neither). ⚠️ `seq` may have gaps (backfilled after the close). ⚠️ No 09:25 auction print — never take an open price here. Page with `after_seq` = previous `next_after_seq` until null (a day is ≤3 pages at the default limit). Retention 60 trading days (`retention_from`); coverage starts 2026-06-01.Not declaredNot declaredNot declaredNot declared
trading_daysTrading days that have a hot-stock board, newest first, as `{dates:[...]}` (China time). **Use this to drive a `date` walk over `hot` — never iterate calendar days**: a weekend `hot` date returns a cumulative window that re-counts notes, whereas these days tile with no overlap and all have data. `days` looks back in CALENDAR days (7-365); coverage begins 2026-02-25. ⚠️ A coverage index for the research board, NOT an exchange calendar — use `market_status` for whether the market is open.Not declaredNot declaredNot declaredNot declared
usageThis key's credit usage for the current month: `used`, `base_cap`, `base_remaining`, `pack_remaining` (purchased top-ups, spent only after the base allowance runs out), `total_remaining`, plus a 6-month history in `monthly`. Use it to decide whether a wide sweep is affordable before starting one, and to tell a 429 that means 'out of credits' apart from one that means 'too fast' — the first will not clear by waiting. ⚠️ **Consumer keys only** (`kind=user`): partner and agent tokens are billed per contract and get 403 here — `whoami` reports quota for those instead. Checking usage is itself free and never cached.Not declaredNot declaredNot declaredNot declared
viewsKEY FIELDS (per item): `report.title` (nullable, only compact/full; index uses `title`), `source_label` (the institution; no `institution`), `ai_summary`, `content` (+ `content_truncated`), `primary_security.canonical_code` / `.name` / `.recommend_action`, `full_text_available`, `published_at`, `id` (for `report`). PRIMARY research source for stock judgement: analyst views and sell-side reports with research_entity_id attribution — prefer over `news`. Only ~10% of items have `full_text_available=true`, so an all-false page says nothing; pass `report_only=true` for the report stream, then `report` each. Default `fields=compact` (summary + ~300 chars); `fields=index` (~150 B/row) for a first pass over many rows, `full` only when needed. A+H: pass the A-share canonical for consolidated coverage.Not declaredNot declaredNot declaredNot declared
views_getFetch one analyst view by id — the read half of list→read, for when a `views` list row is not enough. Returns that view's complete body (never truncated, regardless of the `fields` mode the list was fetched with), its AI summary, attached images, and tagged securities. Use it on any id from `views`; that includes the ~90% of the feed that are plain analyst opinion rather than archived reports. ⚠️ **This is not `report`.** `report` returns the parsed long-form text of an archived sell-side PDF and only works where `full_text_available=true`; this returns the view record itself and works on every id. If the list row already carries what you need, do not call this — a `views` page at `fields=full` has the body inline. Free, and takes no scope of its own.Not declaredNot declaredNot declaredNot declared
whoamiWhat this API key can do: `kind` (user / partner / agent), `scopes`, rate limit, monthly quota, `cli_release` (latest / minimum_supported — is this build behind), `endpoints` (entry-point policy), and `operations.available` / `operations.unavailable` (with `missing_scopes_any` or `not_allowed_for_kind`). Call this **before concluding a capability does not exist** — a missing tool or a 403 is usually an ungranted scope. Advisory (server is the authority). Free, cached 5 minutes.Not declaredNot declaredNot declaredNot declared

Details

Registry status
Active
Published
Nov 15, 2026
Updated
Nov 15, 2026
Categories (suggested)
Finance and paymentsSuggested categories are assigned automatically from the registry text and may be wrong. Verified operators can correct them.
GDPR
Not declared

Are you the operator? Claim this listing