China Dividend Value Screener MCP avatar

China Dividend Value Screener MCP

Pricing

from $30.00 / 1,000 mcp screen call — dividend / value screenings

Go to Apify Store
China Dividend Value Screener MCP

China Dividend Value Screener MCP

A-share dividend + value MCP: TTM/plan dividend yield, PE/PB, ROE and market-cap filters; after-tax yield, A-vs-H tax comparison, dividend-growth quality screen, payout-sustainability rule. 6 tools (describe_fields free). Free Eastmoney data, no API key. Love it? Ko-fi https://ko-fi.com/panstories

Pricing

from $30.00 / 1,000 mcp screen call — dividend / value screenings

Rating

0.0

(0)

Developer

Neeen Ja

Neeen Ja

Maintained by Community

Actor stats

0

Bookmarked

1

Total users

0

Monthly active users

6 days ago

Last modified

Features

Limited permissionsSingle-tenantServer interface

Share

English · China Dividend Value Screener MCP

Current version: 1.5.0

A-share dividend + value screening over MCP — dividend yield, PE/PB, ROE and market-cap filters on free Eastmoney data. No API key, no upstream fees, no data vendor contract.

6 tools · remote Streamable HTTP · Proprietary

Languages: English · 简体中文 · 繁體中文


Why this exists

Screening Chinese A-shares for a dividend-value strategy normally means either paying a data vendor or hand-scraping data.eastmoney.com. This server packages the Eastmoney data-centre reports behind six typed MCP tools, so an agent can answer questions like:

  • "Show me A-shares yielding over 6% with PE under 10 and ROE above 12%."
  • "Has China Merchants Bank paid a dividend every year?"
  • "What is ICBC trading at right now?"

The upstream endpoint is public and free. The cost of running this server is the compute, not the data.

Tools

ToolBillableWhat it does
screen_dividend_stocks$0.03Screen by dividend yield floor + PE ceiling + PB ceiling + ROE floor + market-cap floor, and (v1.5.0) a payout-sustainability rule: a dividend above 100% of reported earnings was funded by asset sales and is excluded. Joins the dividend-plan register with live valuation and the latest ROE.
screen_value_stocks$0.03Classic value screen: low PE_TTM, low PB, positive operating cash flow, market-cap floor.
screen_quality_dividend_growers$0.03Quality dividend-growth screen — finds A-shares that earn their dividend: positive operating cash flow + minimum ROE + a minimum streak of consecutive paying years + a minimum 3-year DPS CAGR + payout ≤ 100% of reported earnings, on top of the value gates. Returns a composite 0–100 rank score.
get_dividend_profile$0.01One stock: recent payout plans (yield, pre-tax amount per 10 shares, ex-dividend date, implementation status), per-plan payout ratio, consecutive dividend years, current valuation and ROE.
get_stock_valuation$0.01One stock: close, PE_TTM, PB_MRQ, PS_TTM, operating cash-flow ratio, market cap, latest ROE.
describe_fieldsfreeField provenance: which upstream report and column each number comes from, whether it is reported or derived, its unit, and the known caveat. Never billed — provenance is not a paid feature.

initialize and tools/list are always free — an agent must be able to connect and enumerate tools before spending anything.

Tool parameters

screen_dividend_stocks

ParamDefaultMeaning
minDividendYieldPct5Minimum dividend yield, in percent
maxPE15Maximum PE_TTM
maxPB3Maximum PB_MRQ
minRoePctnullMinimum ROE, in percent (null = no ROE filter)
minMktcapYi30Minimum market cap in 億 CNY (100M CNY)
annualOnlyfalseKeep only annual-report plans, drop interim dividends
implementedOnlyfalseKeep only plans already implemented, drop proposals
excludeSTtrueDrop ST / *ST / delisting-risk names
excludeBankfalseDrop banks, insurers, brokers
sinceYearlast yearEarliest fiscal year of dividend plans
yieldBasisplanplan = the exchange's yield for one plan (v1.0 behaviour). ttm = every implemented cash dividend with an ex-date in the trailing 12 months, ÷ current close
limit20Max rows
holdingYears1A-share holding period in years for the differential dividend tax (≥1y exempt, 1m–1y 10%, <1m 20%)
taxResidencycn-residentInvestor tax residency. cn-resident-unconfirmed returns after-tax as null instead of guessing
compareHSharefalseAdd an h_share block comparing the A-share against its HK-listed twin for dual-listed names
routehkconnectRoute assumed for the H leg: Stock Connect (港股通) vs an offshore broker
fxHkdCnyliveOverride the live HKD→CNY rate used for the H comparison
codesnullWatchlist mode. An explicit list of 6-digit A-share codes. The screen is restricted to exactly those codes — the market-wide dividend scan is skipped, the value/yield filters are NOT applied (you chose the names), and every requested code comes back even if it has no dividend plan (n/a fields). ST exclusion still applies.
livePricesfalseForce a live A-share quote even when no H comparison is requested. codes mode always prices live.

Under yieldBasis: "ttm" the headline dividend_yield_pct becomes the trailing aggregate, and each row also carries ttm_dps, ttm_dividend_yield_pct and ttm_spans_fiscal_years. Measured effect: 索菲亚 reads 7.16% on one plan and 10.68% on TTM, because it pays twice a year.

screen_value_stocks — maxPE (10), maxPB (2), minMktcapYi (30), requirePositivePCF (true), excludeST (true), excludeBank (false), limit (20), tradeDate (optional).

screen_quality_dividend_growers — minYieldPct (2.5), minRoePct (12), maxPE (20), maxPB (3), minMktcapYi (100), requirePositivePCF (true), minConsecutiveYears (5), minDpsCagrPct (5), excludeST (true), excludeBank (false), limit (34), tradeDate (optional). The forward yield is the latest fiscal-year DPS ÷ close; the 3-year DPS CAGR is computed from the per-year dividend register. Banks are flagged separately because their operating-cash-flow ratio is structurally distorted.

get_stock_valuation — code (6-digit A-share code, e.g. 601398).

get_dividend_profile — code, years (5).

After-tax yield & A-vs-H comparison (v1.2)

screen_dividend_stocks now returns an after-tax yield and can compare an A-share against its Hong Kong twin. The headline yield is what the company pays; the after-tax figure is what a mainland investor actually keeps — and the two are not the same.

Each row gains after_tax_yield_pct, tax_rate_pct, tax_rule, tax_note. With compareHShare: true, dual-listed rows also gain an h_share block: the CNY-normalised H yield (pre_tax_yield_pct), its after-tax figure, the ah_premium_pct (A close ÷ CNY-normalised H close), and better_after_tax (A / H / tie).

The H yield is put on a CNY-normalised basis — the same RMB dividend-per-share (a dual-listed company pays one RMB dividend on both share classes) divided by the HKD price × a live HKD→CNY rate — so it is directly comparable to the A-share CNY yield.

The honest finding this surfaces:

  • For a mainland tax resident who files compliantly, every offshore route converges on ~20% (source withholding is only a credit against the resident 20%), so the "Singapore broker pays only 10%" edge is a non-filing risk, not a saving. Stock Connect (港股通) is a closed loop — the withheld amount is final, no annual filing.
  • The reason an H yield looks higher is usually the A/H valuation discount, not tax. So compare after-tax, CNY-normalised yields.
  • Live example (2026-10-22): 中国移动 A after-tax 4.76% vs H after-tax 5.14% — H still wins despite the 28% red-chip tax, because H trades near ¥66 while A trades near ¥99.

Share class (mainland H-share vs overseas red-chip vs HK-local vs foreign) is curated in a maintained table; a code that is not classified returns n/a, never a guessed rate. The tax model is a dated deterministic snapshot — not tax or investment advice; confirm cross-border residency with a tax professional.

Watchlist mode & same-timestamp A/H premium (v1.3.0)

Two things were wrong before v1.3.0, and both produced numbers you should not have trusted:

  1. The premium mixed two vintages. The H leg used a live intraday H price while the A leg used the previous close. A premium built from two different timestamps is not a premium — it is noise. Now both legs are live (A-share and H-share, both from Tencent gtimg), so ah_premium_pct is same-timestamp. Each row reports a_price_basis (live or eod_close) so you can see which price the A leg used; live_price, live_change_pct and live_ts are also returned.
  2. You could not screen your own list. screen_dividend_stocks only ever started from a market-wide dividend scan, so a name you already knew about could not be priced on its own terms. codes fixes that — point it at a list of 6-digit codes and it returns exactly those, priced live, with the filters skipped and ST names still excluded.

Also in v1.3.0: 中国海油 (600938) is classified as a red-chip (CNOOC Ltd is incorporated in Hong Kong), so its H leg takes the 28% Stock Connect bucket rather than 20%. And the FX source is now intraday — Sina fx_shkdcny first, open.er-api.com as fallback — with a sanity band of 0.75–1.05 CNY per HKD; an out-of-band value is refused rather than used.

FX convention, read this before comparing any premium. fx_hkd_cny is the market rate for 1 HKD in CNY (≈0.85). It is not the PBOC 中间价 and not the inverse CNY→HKD rate (≈1.17). Some vendors quote HKD as CNY per 100 HKD, others quote the inverse; multiplying the H price by an inverted rate inflates the premium by ~1.4× and can turn a 139% premium into a plausible-looking 9%. Always multiply the HKD price by this rate.

Quality dividend-growth screen (v1.4.0, corrected in v1.4.1, payout gate in v1.5.0)

New tool screen_quality_dividend_growers. The dividend screen answers "who pays the most"; this one answers "who can keep paying — and grow it." It starts from a market-wide value/cash-flow base (PE ≤ 20, PB ≤ 3, positive operating cash flow, market cap ≥ 100億), then layers on the quality bars:

  • ROE ≥ 12% (latest annual weighted-average),
  • ≥ 5 consecutive years of implemented cash dividends,
  • ≥ 5% 3-year DPS CAGR (the payout is actually growing, not flat),
  • forward yield ≥ 2.5% — the latest complete fiscal-year DPS ÷ close, i.e. every implemented payment of that year, interim plus final — what a buyer today captures, not an old announcement-price ratio,
  • payout ≤ 100% of reported earnings (v1.5.0) — the dividend must be earned, not liquidated. See below.

It returns a composite 0–100 score (min-max of yield / ROE / CAGR across the survivors) so the set is ranked rather than just filtered. Two honest caveats baked into the notes: (1) a very high DPS CAGR is often a low-base / recovery effect, not durable compounding; (2) for banks the positive-cash-flow gate is meaningless because deposit flows distort the ratio — judge them on capital adequacy instead. The share-class table was extended with four A+H names that the quality screen surfaces (中国铝业, 海尔智家, 海信家电, 青岛啤酒), so the existing A-vs-H overlay now covers them too.

v1.4.1 fix — the forward yield now counts the whole year. v1.4.0 took the December (final) dividend row alone as the year's DPS. Many A-share payers split the year — 格力电器 paid a 1.0 Q3 dividend plus a 2.0 final for FY2025 (3.0 in total, independently confirmed by the trailing-12-month window) — so the December-only read understated their yield by a third or more. Because this screen gates on the forward yield, that silently ejected those names from the universe: 中国铝业 scored 1.70% on the final alone and failed the 2.5% gate, when the full fiscal year is 3.13%. The payout ratio and the 3-year CAGR are now computed on complete fiscal years too, so a half-paid current year (an interim with no December dividend yet) can never be mistaken for a run-rate or a growth base.

v1.5.0 — the dividend must be EARNED (payout sustainability). A yield is only income if the business earned it. When a company pays out more than its reported earnings (payout > 100%), the cash came from somewhere else: disposals of long-term assets, reserve drawdown, or debt. That headline yield looks generous in the year it happens and then evaporates — it is a liquidation in instalments, not a dividend. The quality screen now excludes every name whose payout exceeds maxPayoutRatioPct (default 100%) of that fiscal year's reported EPS, and every name that paid a cash dividend in a year it lost money (an unbounded payout). Three honesty details: a name whose EPS could not be sourced is kept but flagged payout_verifiable: false — unverifiable is not the same as safe; the payout ratio uses the full-year DPS (interim + final) against that year's annual EPS, and the denominator is cross-checked against the financial-indicator report; and every cut name is listed in payoutExcluded (code, name, fiscal year, payout, forward yield) so the exclusion is auditable rather than silent. payout_ratio_3y_avg_pct separates a chronic over-payer from a one-off special dividend.

Data source

Three free, keyless Eastmoney reports, all served from the same public endpoint that powers data.eastmoney.com:

ReportUsed for
RPT_VALUEANALYSIS_DETPE_TTM, PB_MRQ, PCF, PS_TTM, close price, total market cap
RPT_SHAREBONUS_DETDividend plans, DIVIDENT_RATIO (the exchange's own dividend yield), amount per 10 shares, ex-dividend date, implementation status
RPT_LICO_FN_CPDWeighted-average ROE, EPS, BPS

Data semantics you should know

  • Interim dividends are not annualised. A plan filed against a 2026-06-30 period is a half-year yield. Use annualOnly: true when you want annual-report plans only. Every row carries a report_period label (2025A, 2026H1, …) so the basis is never hidden.
  • ROE basis is labelled. The server prefers the newest annual report and falls back to the newest interim one; each row reports roe_basis (annual / interim) and roe_period, because a half-year ROE must not be silently compared against a 10% annual hurdle.
  • Missing data stays missing. Anything that cannot be sourced is returned as null, never defaulted to 0 — a fabricated 0% yield is worse than an absent one.
  • One trading day per screen. tradeDate in the response tells you exactly which session the valuation snapshot is from.
  • DIVIDENT_RATIO is one plan, not the stock. It is the exchange's own yield for a single distribution, measured against the price on that plan's announcement date — it ignores a second payment in the same year and uses a historical price. get_dividend_profile therefore also returns a TTM block (ttm_dps, ttm_dividend_yield_pct, ttm_window, ttm_spans_fiscal_years): the sum of implemented cash dividends with an ex-date in the trailing 12 months, divided by the current close. Pre-tax. Amounts are quoted per 10 shares and divided by 10.
  • Consecutive dividend years counts only money that was actually paid. 预披露 and 董事会决议通过 have no ex-date and may never be paid, so they no longer extend the streak. When the streak runs to the edge of the retrieved history, consecutive_years_is_lower_bound is true and the number is a floor.
  • A dividend must be earned (v1.5.0). Both dividend screens exclude a name whose payout exceeds 100% of that fiscal year's reported earnings, and any name that paid a cash dividend in a year it lost money: the cash came out of the balance sheet, so the yield is a liquidation in instalments that cannot repeat. The payout is measured on the newest complete fiscal year (interim + final DPS ÷ that year's annual EPS) and is reported per row as payout_ratio_pct / payout_basis / payout_year / payout_eps. Every excluded name is returned in payoutExcluded. Pass maxPayoutRatioPct: null to switch the rule off, or a higher number to relax it; the rule is not applied in codes (watchlist) mode.

Connect

https://<your-actor-run-id>.apify.actor/mcp

Send Authorization: Bearer <APIFY_TOKEN>. The endpoint speaks Streamable HTTP; POST JSON-RPC 2.0 messages (initialize, tools/list, tools/call).

Self-hosted deployment (on request)

The recommended way to use this server is the hosted endpoint above — zero setup, you call the tools and consume the screening results. The source code is private and available on request; self-hosted deployment of the server can be arranged by contacting PanStories. There is no public npx package and no public Git repository.

Example response

A-share dividend + value screen · trading day 2026-09-30
Filters: yield ≥ 5% · PE_TTM ≤ 20 · PB ≤ 3 · ROE ≥ n/a% · market cap ≥ 50億 CNY
Scanned 60 dividend plans from 2025 onwards, showing 3.
1. 601216 君正集团 (化学原料) yield 7.95% [2025A] · PE 12.5 · PB 1.39 · ROE 11.78% (2025-12-31) · mktcap 383.09億
2. 600861 北京人力 (专业服务) yield 7.23% [2025A] · PE 9.47 · PB 1.02 · ROE 16.99% (2025-12-31) · mktcap 72.63億
3. 002572 索菲亚 (家居用品) yield 7.16% [2025A] · PE 11.55 · PB 1.11 · ROE 12.27% (2025-12-31) · mktcap 74.35億

Every tool returns both a human-readable summary (content) and machine-readable data (structuredContent), so agents never have to parse prose.

Pricing

EventPriceTools
mcp-screen$0.03screen_dividend_stocks, screen_value_stocks
mcp-profile$0.01get_dividend_profile, get_stock_valuation

Screens cost more because they fan out to three upstream reports; single-stock lookups take two. Discovery is free, and describe_fields is free on purpose. Set maxTotalChargeUsd on the run to cap autonomous agent spend.

Price update (test): mcp-screen rises from $0.02 → $0.03 effective 2026-11-03 (14-day notice; existing users keep $0.02 until then). mcp-profile stays $0.01. The v1.1.0 release itself did not change prices.

Limitations

  • A-shares drive the screen. The screener seeds on Shanghai / Shenzhen / Beijing listings; compareHShare adds an after-tax comparison against the Hong Kong twin for dual-listed names — it is not a full HK screener. For US / Taiwan listings see the companion mcp-stock-analyst server.
  • The tax model is a dated snapshot. Dividend-tax rules (v1.2 model, dated 2026-10-22) are encoded deterministically and are not tax advice; cross-border residency and treaty cases need a tax professional. Share class is curated in a maintained table — an unclassified code returns n/a, never a guessed rate.
  • Fundamentals are end-of-day; prices can be live. PE / PB / ROE / market cap come from the daily valuation snapshot. Prices (A-share and H-share) are live gtimg quotes when requested, and a_price_basis / live_ts tell you which price a row used.
  • Not investment advice. This server screens public data. It does not know your position, your horizon or your tax situation.

Source & licence

Proprietary © PanStories. Source is available on request; the repository remains private. This server is offered as a hosted service (pay-per-event on Apify) — you consume the screening results, not the code.

Privacy — read-only and stateless; no accounts, no cookies, no trackers, no personal data collected. See the Privacy Policy below.


简体中文

当前版本:1.5.0

这是什么

A 股高股息 + 价值筛选 MCP 服务。用免费东方财富数据做股息率、PE/PB、ROE、市值门槛筛选,无需 API key,零上游数据成本。

六个工具

工具计费作用
screen_dividend_stocks$0.03按股息率下限 + PE 上限 + PB 上限 + ROE 下限 + 市值下限筛选;并强制派息可持续性规则(v1.5.0):派息率超过当年报表盈利 100% 的标的被剔除(那笔钱来自变卖资产);把分红方案台账与实时估值、最新 ROE 三张表做联结。可选 yieldBasis="ttm" 改用滚动 12 个月真实派息合计口径
screen_value_stocks$0.03经典价值筛选:低 PE_TTM、低 PB、经营现金流为正、市值门槛
screen_quality_dividend_growers$0.03高质量股息成长筛选:找「配得上」股息的 A 股——经营现金流为正 + ROE 下限 + 连续分红年数下限 + 3 年每股派息 CAGR 下限,再叠加估值门槛;返回 0–100 综合评分
get_dividend_profile$0.01单只股票:历次分红方案(股息率、每 10 股税前派息、除权日、实施进度)、单次派息率、连续分红年数、TTM 真实股息率、当前估值与 ROE
get_stock_valuation$0.01单只股票:收盘价、PE_TTM、PB_MRQ、PS_TTM、现金流市销比、市值、最新 ROE
describe_fields免费字段溯源:每个数字来自哪张报表哪一列、是实测值还是推导值、单位、以及已知口径坑。永不计费

initialize 与 tools/list 永久免费 —— 智能体必须先能连上并列出工具,才谈得上消费。

税后股息率 & A/H 对比(v1.2)

screen_dividend_stocks 现在额外返回税后股息率,并可把 A 股和它的港股「孪生」做税后对比。表观股息率是公司派多少,税后才是大陆投资者实际到手——两者不是一回事。

每行新增 after_tax_yield_pct、tax_rate_pct、tax_rule、tax_note;传 compareHShare: true 时,A+H 两地上市的行还会多一个 h_share 区块:CNY 归一化后的 H 股股息率(pre_tax_yield_pct)、其税后值、ah_premium_pct(A 股价 ÷ CNY 归一化 H 股价)、以及 better_after_tax(A/H/tie)。H 股用同一笔人民币每股派息(两地上市派的是同一条人民币股息)÷(HKD 股价 × 实时 HKD→CNY)折算,可直接与 A 股 CNY 股息率比较。

新增参数:holdingYears(默认 1,A 股持股期限差别化税)、taxResidency(默认 cn-resident;cn-resident-unconfirmed 时税后返回 null 不猜)、compareHShare(默认 false)、route(默认 hkconnect)、fxHkdCny(覆盖实时汇率)。

它讲清的一个诚实结论:对合规申报的大陆税务居民,任何境外路径最终都收敛到约 20%(源头预扣只是抵免项),所谓「新加坡券商只扣 10%」的优势本质是不申报的风险折价,不是真省税;港股通是闭环,扣完即终局。而 H 股股息率看起来更高,通常是 A/H 估值折价造成的,不是税。所以要比就比税后、CNY 归一化的股息率。实测(2026-10-22):中国移动 A 税后 4.76% vs H 税后 5.14%——即便红筹股被扣 28%,H 仍胜出,因为 H 约 ¥66 而 A 约 ¥99。

自选股模式 & 同时点 A/H 溢价(v1.3.0)

v1.3.0 之前有两个地方会给出不该被信任的数字,都已修掉:

  1. 溢价混了两个时点。 H 腿用的是当日实时价,A 腿用的是前收盘价。两个不同时点算出来的溢价不是溢价,是噪音。现在两腿都是实时价(A 股、H 股均取腾讯 gtimg),ah_premium_pct 是同时点的。每行新增 a_price_basis(live / eod_close)标明 A 腿用了哪个价;同时返回 live_price、live_change_pct、live_ts。
  2. 没法筛自己的清单。 screen_dividend_stocks 只能从全市场分红扫描起步,你已经知道的票无法单独定价。codes 解决这件事——传一组 6 位代码,只返回这些票,按实时价定价,不过滤值/收益率门槛(票是你自己选的),没有分红方案的票也会返回(相关字段为 n/a)。ST 剔除依然生效。

v1.3.0 另外两项:中国海油(600938)归类为红筹(CNOOC Ltd 注册于香港),H 腿走港股通 28% 档而非 20%;汇率改为盘中实时——新浪 fx_shkdcny 优先、open.er-api.com 兜底,并设 0.75–1.05 的合理区间,越界直接拒绝而不是照用。

汇率口径,看溢价前必读:fx_hkd_cny 是1 港币兑多少人民币的市场价(≈0.85),不是央行中间价,也不是反向的人民币兑港币(≈1.17)。有的数据商把港币报成「100 港币兑多少人民币」,有的直接给反向报价;用反向汇率乘 H 股价,溢价会被放大约 1.4 倍,139% 的溢价会变成看着挺合理的 9%。用这个汇率乘 HKD 股价就对了。

股类(H 股/红筹/香港本地/外资)由人工维护的分类表判定;未收录的代码返回 n/a,绝不猜。税务模型是带日期的确定性快照,不构成税务/投资建议;跨境税务身份请咨询专业税务师。

高质量股息成长筛选(v1.4.0,v1.4.1 修正,v1.5.0 加派息可持续性门槛)

新增工具 screen_quality_dividend_growers。股息率筛选回答「谁派得最多」;这个工具回答「谁配得上长期派、而且还在增长」。它先取全市场价值/现金流底座(PE ≤ 20、PB ≤ 3、经营现金流为正、市值 ≥ 100 亿),再叠加质量门槛:

  • ROE ≥ 12%(最新年报加权);
  • 连续分红 ≥ 5 年(只算已实施的现金分红);
  • 3 年每股派息 CAGR ≥ 5%(派息在增长,而非持平);
  • 前瞻股息率 ≥ 2.5%(最新完整财年 DPS ÷ 收盘价——即该财年全部已实施派息,中期+年末都算——买今天的人实际拿到的,不是旧公告价算的);
  • 派息率 ≤ 报表盈利的 100%(v1.5.0)——股息必须是「赚出来的」,而不是变卖资产变出来的,见下。

它返回一个 0–100 综合评分(在幸存者中对收益率/ROE/CAGR 做 min-max 归一化),让结果按质量排序而非仅过滤。两点诚实提醒写进了 notes:(1) 很高的 DPS CAGR 往往是低基数/恢复效应,不是可持续复利;(2) 对银行来说,现金流为正这道门槛没有意义——存款流动会扭曲该比率,应按资本充足率而非 PCF 评判。分类表也补了质量筛选新出现的 4 个 A+H 标的(中国铝业、海尔智家、海信家电、青岛啤酒),所以现有的 A/H 税后对比现在也覆盖它们。

v1.4.1 修正——前瞻股息率现在算整年。 v1.4.0 只取 12 月(年末)那一行派息当作全年 DPS。大量 A 股是「一年两派」——格力电器 FY2025 派了 1.0(三季度)+ 2.0(年末)= 3.0(滚动 12 个月窗口独立印证同为 3.0)——只取年末会低估三分之一以上。由于这道筛选以前瞻股息率为门槛,那些名字被静默踢出了宇宙:中国铝业按年末单行算只有 1.70%,够不到 2.5% 门槛,而其完整财年实际是 3.13%。派息率与 3 年 CAGR 现在也只用完整财年计算,半个已过年份(只派了中期、年末还没派)再也不会被误当年化基数或增长基数。

v1.5.0——股息必须赚出来(派息可持续性门槛)。 只有公司赚到的股息才算收益。当一家公司派息超过当年报表盈利(派息率 > 100%)时,这笔钱来自别处:变卖长期资产、动用储备或举债。这种「高股息」只在当年好看,随后就消失——它是分期清算,不是分红。质量筛选现在剔除派息率超过 maxPayoutRatioPct(默认 100%,即该财年报表 EPS 的 100%)的标的,也剔除在亏损财年仍然派现金股息的标的(派息无上限)。三点诚实说明:取不到 EPS 的标的会保留但标记 payout_verifiable: false——「无法验证」不等于「安全」;派息率的分母用全年 DPS(中期+年末)对该年年报 EPS,并与财务指标报告交叉核对;每只被剔除的标的都会列在 payoutExcluded(代码、名称、财年、派息率、前瞻股息率)里,剔除可审计而不是静默。payout_ratio_3y_avg_pct 用来区分「长期超额派息」与「一次性特别分红」。

必须知道的数据口径

  • 中期分红不做年化。 归属 2026-06-30 的方案就是半年口径的股息率。只要年报刊方案,把 annualOnly 设为 true。每行都带 report_period 标签(2025A/2026H1…),口径从不含糊。
  • ROE 口径明示。 优先取最新年报 ROE,取不到才退回最新中报;每行都给出 roe_basis(annual/interim)与 roe_period —— 半年 ROE 不能悄悄拿去和「年化 10%」的门槛比。
  • 取不到就是 null。 任何无法取到的数字一律返回 null,绝不默认为 0。编出来的 0% 股息率比缺失更害人。
  • 每次筛选只对应一个交易日。 响应里的 tradeDate 明确告诉你估值快照取自哪一天。
  • DIVIDENT_RATIO 是「单笔方案」不是「这只股票」。 它是交易所按该方案公告日的股价算的单次股息率:同年第二次派息不算,用的还是旧股价。因此 get_dividend_profile 额外返回 TTM 区块(ttm_dps、ttm_dividend_yield_pct、ttm_window、ttm_spans_fiscal_years)——滚动 12 个月内已实施现金分红合计 ÷ 当前收盘价(税前;派息额按每 10 股计,需 ÷10)。实测差异:索菲亚单笔口径 7.16%,TTM 口径 10.68%(一年两派)。
  • 连续分红年数只算真的发了钱的分红。 预披露、董事会决议通过 没有除权日、也可能最终不发,不再延长连续年数;若连续年数一直顶到已取历史的边界,consecutive_years_is_lower_bound 为 true,该数字是下界。
  • 股息必须赚出来(v1.5.0)。 两个分红筛选都会剔除派息率超过当年报表盈利 100% 的标的,以及亏损财年仍派现金股息的标的——这笔钱来自资产负债表,是「分期清算」而不是收入,不可能重复。派息率按最新完整财年计算(中期+年末 DPS ÷ 该年年报 EPS),每行返回 payout_ratio_pct / payout_basis / payout_year / payout_eps;被剔除的标的全部列在 payoutExcluded 里,可审计而非静默。传 maxPayoutRatioPct: null 可关闭该规则,传更大的数字可放宽;codes(自选股)模式不套用该规则。

接入

  • Apify 托管(推荐):https://<run-id>.apify.actor/mcp,带 Authorization: Bearer <APIFY_TOKEN>,Streamable HTTP 协议,POST JSON-RPC 2.0。
  • 自托管部署(按需提供):推荐用上面的托管端点,零配置、直接消费筛选结果。源码为私有、可按需提供;自托管部署请联系 PanStories。无公开 npx 包、无公开 Git 仓库。

计费

筛选类工具(screen_*)$0.03/次,单只查询(get_*)$0.01/次。筛选贵一倍是因为它要扇出到三张上游报表,单只查询只用两张。describe_fields 永久免费。

价格更新(测试):mcp-screen 自 2026-11-03 起由 $0.02 → $0.03(14 天通知期;存量用户在此之前仍按 $0.02)。mcp-profile 维持 $0.01。

限制

  • 筛选以 A 股为主(沪/深/京);compareHShare 只是对 A+H 两地上市标的加一个税后对比,并非完整的港股筛选器。美股/台股请用配套的 mcp-stock-analyst。
  • 税务模型是带日期的快照(v1.2,2026-10-22),确定性编码,不构成税务建议;跨境身份与税收协定需咨询专业税务师。股类为人工维护——未收录代码返回 n/a,绝不猜。
  • 基本面为收盘口径,价格可实时。 PE / PB / ROE / 市值取自每日估值快照;价格(A 股与 H 股)在请求时取 gtimg 实时报价,a_price_basis / live_ts 标明该行用的是哪个价。
  • 不构成投资建议。本服务只做公开数据的筛选,不知道你的持仓、期限与税负。

源码与许可

专有(Proprietary)© PanStories。源码可按需提供;仓库保持私有。本服务以托管形式提供(Apify 按事件计费)—— 你消费的是筛选结果,不是代码。

隐私 / Privacy — 只读、无状态;无账号、无 Cookie、无追踪,不收集任何个人信息。详见下文 隐私政策。


繁體中文

當前版本:1.5.0

這是什麼

A 股高股息 + 價值篩選 MCP 服務。以免費東方財富資料做股息率、PE/PB、ROE、市值門檻篩選,無需 API key、零上游資料成本。

六個工具

工具計費作用
screen_dividend_stocks$0.03依股息率下限 + PE 上限 + PB 上限 + ROE 下限 + 市值下限篩選;並強制配息可持續性規則(v1.5.0):配息率超過當年財報盈餘 100% 的標的剔除(那筆錢來自變賣資產);將分紅方案台帳與即時估值、最新 ROE 三張表做聯結。可選 yieldBasis="ttm" 改用滾動 12 個月實際派息合計口徑
screen_value_stocks$0.03經典價值篩選:低 PE_TTM、低 PB、營業現金流為正、市值門檻
screen_quality_dividend_growers$0.03高品質股息成長篩選:找「配得上」股息的 A 股——營業現金流為正 + ROE 下限 + 連續分紅年數下限 + 3 年每股派息 CAGR 下限,再疊加估值門檻;回傳 0–100 綜合評分
get_dividend_profile$0.01單一個股:歷次分紅方案(股息率、每 10 股稅前派息、除權日、實施進度)、單次派息率、連續分紅年數、TTM 真實股息率、當前估值與 ROE
get_stock_valuation$0.01單一個股:收盤價、PE_TTM、PB_MRQ、PS_TTM、現金流市銷比、市值、最新 ROE
describe_fields免費欄位溯源:每個數字來自哪張報表哪一欄、是實測值還是推導值、單位,以及已知口徑陷阱。永不計費

initialize 與 tools/list 永久免費 —— 智慧體必須先能連上並列出工具,才談得上消費。

稅後股息率 & A/H 比較(v1.2)

screen_dividend_stocks 現在額外回傳稅後股息率,並可把 A 股與其港股「雙生」做稅後比較。每列新增 after_tax_yield_pct、tax_rate_pct、tax_rule、tax_note;傳 compareHShare: true 時,A+H 兩地上市列另增 h_share 區塊(CNY 正規化 H 股息率、稅後值、ah_premium_pct、better_after_tax)。H 股以同一筆人民幣每股派息 ÷(HKD 股價 × 即時 HKD→CNY)正規化,可直接與 A 股 CNY 股息率相比。新增參數:holdingYears、taxResidency、compareHShare、route、fxHkdCny。

誠實結論:對合規申報的大陸稅務居民,任何境外路徑最終都收斂到約 20%,「新加坡券商只扣 10%」本質是不申報的風險折價;港股通是閉環。H 股股息率看起來更高多是 A/H 估值折價所致,非稅。要比就比稅後、CNY 正規化股息率。實測(2026-10-22):中國移動 A 稅後 4.76% vs H 稅後 5.14%——紅籌被扣 28% 仍 H 勝出。股類為人工維護,未收錄回傳 n/a 絕不猜;模型為帶日期快照,不構成稅務/投資建議。

自選股模式 & 同時點 A/H 溢價(v1.3.0)

v1.3.0 之前有兩處會給出不該被信任的數字,皆已修正:

  1. 溢價混了兩個時點。 H 腿是當日即時價,A 腿卻是前收盤價。兩個時點算出的溢價不是溢價,是雜訊。現在兩腿皆為即時價(A 股、H 股均取騰訊 gtimg),ah_premium_pct 為同時點。每列新增 a_price_basis(live/eod_close)標示 A 腿用價,另回傳 live_price、live_change_pct、live_ts。
  2. 無法篩自己的清單。 screen_dividend_stocks 過往只能從全市場分紅掃描起步。codes 讓你傳一組 6 位代碼,只回傳這些標的,按即時價定價,不套用值/收益率門檻(標的是你自己選的),無分紅方案的標的亦回傳(相關欄位 n/a)。ST 剔除仍生效。

v1.3.0 另兩項:中國海油(600938)歸類為紅籌(CNOOC Ltd 註冊於香港),H 腿走港股通 28% 檔;匯率改為盤中即時——新浪 fx_shkdcny 優先、open.er-api.com 備援,並設 0.75–1.05 合理區間,越界直接拒用。

匯率口徑:fx_hkd_cny 是1 港幣兌多少人民幣的市場價(≈0.85),不是央行中間價,也不是反向的人民幣兌港幣(≈1.17)。部分資料商以「100 港幣兌多少人民幣」報價或直接給反向報價;用反向匯率乘 H 股價會把溢價放大约 1.4 倍,139% 溢價會變成看似合理的 9%。

高品質股息成長篩選(v1.4.0,v1.4.1 修正,v1.5.0 加入配息可持續性門檻)

新增工具 screen_quality_dividend_growers。股息率篩選回答「誰派得最多」;此工具回答「誰配得上長期派、而且還在成長」。它先取全市場價值/現金流底座(PE ≤ 20、PB ≤ 3、營業現金流為正、市值 ≥ 100 億),再疊加品質門檻:

  • ROE ≥ 12%(最新年報加權);
  • 連續分紅 ≥ 5 年(只算已實施的現金分紅);
  • 3 年每股派息 CAGR ≥ 5%(派息在成長,而非持平);
  • 前瞻股息率 ≥ 2.5%(最新完整財年 DPS ÷ 收盤價——即該財年全部已實施派息,期中+年末都算——買今天的人實際拿到的,不是舊公告價算的);
  • 配息率 ≤ 財報盈餘的 100%(v1.5.0)——股息必須是「賺出來的」,而不是變賣資產變出來的,見下。

它回傳一個 0–100 綜合評分(在倖存者中對收益率/ROE/CAGR 做 min-max 正規化),讓結果按品質排序而非僅過濾。兩點誠實提醒寫進 notes:(1) 很高的 DPS CAGR 往往是低基數/恢復效應,不是可持續複利;(2) 對銀行來說,現金流為正這道門檻沒有意義——存款流動會扭曲該比率,應按資本充足率而非 PCF 評判。分類表也補了品質篩選新出現的 4 個 A+H 標的(中國鋁業、海爾智家、海信家電、青島啤酒),所以現有的 A/H 稅後比較現在也覆蓋它們。

v1.4.1 修正——前瞻股息率現在計算整年。 v1.4.0 只取 12 月(年末)那一列派息當作全年 DPS。大量 A 股是「一年兩派」——格力電器 FY2025 派了 1.0(三季度)+ 2.0(年末)= 3.0(滾動 12 個月視窗獨立印證同為 3.0)——只取年末會低估三分之一以上。由於這道篩選以前瞻股息率為門檻,那些名稱被靜默踢出了宇宙:中國鋁業按年末單列算只有 1.70%,夠不到 2.5% 門檻,而其完整財年實際是 3.13%。派息率與 3 年 CAGR 現在也只用完整財年計算,半個已過年份(只派了期中、年末還沒派)再也不會被誤當年化基數或成長基數。

v1.5.0——股息必須賺出來(派息可持續性)。 只有公司賺到的股息才算收益。當一家公司派息超過當年財報盈餘(派息率 > 100%)時,這筆錢來自別處:變賣長期資產、動用公積或舉債。這種「高股息」只在當年好看,隨後就消失——它是分期清算,不是股利。品質篩選現在剔除派息率超過 maxPayoutRatioPct(預設 100%,即該財年報表 EPS 的 100%)的標的,也剔除在虧損財年仍然派現金股利的標的(派息無上限)。三點誠實說明:取不到 EPS 的標的會保留但標記 payout_verifiable: false——「無法驗證」不等於「安全」;派息率的分母用全年 DPS(期中+年末)對該年年報 EPS,並與財務指標報告交叉核對;每隻被剔除的標的都列在 payoutExcluded(代碼、名稱、財年、派息率、前瞻股息率)裡,剔除可稽核而不是靜默。payout_ratio_3y_avg_pct 用來區分「長期超額派息」與「一次性特別股利」。

必須知道的資料口徑

  • 期中分紅不做年化。 歸屬 2026-06-30 的方案就是半年口徑的股息率。只要年報方案,請把 annualOnly 設為 true。每一列都帶 report_period 標籤(2025A/2026H1…),口徑從不含糊。
  • ROE 口徑明示。 優先取最新年報 ROE,取不到才退回最新期中報;每一列都給出 roe_basis(annual/interim)與 roe_period —— 半年 ROE 不該悄悄拿去和「年化 10%」的門檻相比。
  • 取不到就是 null。 任何無法取得的數字一律回傳 null,絕不預設為 0。捏造出來的 0% 股息率比缺漏更糟。
  • 每次篩選只對應一個交易日。 回應中的 tradeDate 明確指出估值快照取自哪一天。
  • DIVIDENT_RATIO 是「單一方案」而非「這檔股票」。 它是交易所依該方案公告日股價算出的單次股息率:同一年的第二次派息不計入,用的還是舊股價。因此 get_dividend_profile 額外回傳 TTM 區塊(ttm_dps、ttm_dividend_yield_pct、ttm_window、ttm_spans_fiscal_years)——滾動 12 個月內已實施現金分紅合計 ÷ 當前收盤價(稅前;派息額為每 10 股,需 ÷10)。實測差異:索菲亞單一方案口徑 7.16%,TTM 口徑 10.68%(一年兩次派息)。
  • 連續分紅年數只算真正發出去的錢。 預披露、董事會決議通過 沒有除權日、也可能最終不發,不再延長連續年數;若連續年數頂到已取歷史的邊界,consecutive_years_is_lower_bound 為 true,該數字是下界。
  • 股息必須賺出來(v1.5.0)。 兩個分紅篩選都會剔除配息率超過當年財報盈餘 100% 的標的,以及虧損財年仍配現金股息的標的——這筆錢來自資產負債表,是「分期清算」而不是收入,不可能重複。配息率按最新完整財年計算(期中+年末 DPS ÷ 該年年報 EPS),每一列回傳 payout_ratio_pct / payout_basis / payout_year / payout_eps;被剔除的標的全部列在 payoutExcluded 裡,可稽核而非靜默。傳 maxPayoutRatioPct: null 可關閉該規則,傳更大的數字可放寬;codes(自選股)模式不套用該規則。

接入

  • Apify 託管(推薦):https://<run-id>.apify.actor/mcp,帶 Authorization: Bearer <APIFY_TOKEN>,Streamable HTTP 協定,POST JSON-RPC 2.0。
  • 自託管部署(按需提供):推薦用上面的託管端點,零配置、直接消費篩選結果。原始碼為私有、可依需求提供;自託管部署請聯絡 PanStories。無公開 npx 套件、無公開 Git 倉庫。

計費

篩選類工具(screen_*)$0.03/次,單一個股查詢(get_*)$0.01/次。篩選貴一倍是因為它要扇出到三張上游報表,單一個股查詢只用兩張。describe_fields 永久免費。

價格更新(測試):mcp-screen 自 2026-11-03 起由 $0.02 → $0.03(14 天通知期;存量用戶在此之前仍按 $0.02)。mcp-profile 維持 $0.01。v1.1.0 發布本身未改價。

限制

  • 篩選以 A 股為主(滬/深/京);compareHShare 僅對 A+H 兩地上市標的加一個稅後比較,並非完整港股篩選器。美股/台股請用配套的 mcp-stock-analyst。
  • 稅務模型為帶日期快照(v1.2,2026-10-22),不構成稅務建議;跨境身份與稅收協定請諮詢專業稅務師。股類人工維護——未收錄回傳 n/a,絕不猜。
  • 基本面為收盤口徑,價格可即時。 PE / PB / ROE / 市值取自每日估值快照;價格(A 股與 H 股)在請求時取 gtimg 即時報價,a_price_basis / live_ts 標示該列用的是哪個價。
  • 不構成投資建議。本服務僅就公開資料做篩選,不知道你的持倉、期限與稅負。

原始碼與授權

專有(Proprietary)© PanStories。原始碼可依需求提供;倉庫保持私有。本服務以託管形式提供(Apify 按事件計費)—— 你消費的是篩選結果,不是程式碼。

隱私 / Privacy — 唯讀、無狀態;無帳號、無 Cookie、無追蹤,不收集任何個人資訊。詳見下文 隱私政策。


Privacy Policy

Product: China Dividend Value Screener MCP (MCP server) Operator: PanStories Last updated: 2026-10-09 Effective date: 2026-10-09

This is the authoritative privacy policy for this hosted service. It is reproduced in full here because the source repository is private; the identical text also ships as PRIVACY.md in the repository.

1. Summary (TL;DR)

  • The Service is a read-only MCP server. Its tools only read public A-share market data on free Eastmoney endpoints to screen for dividend/value metrics. They never modify, delete, or act on your systems.
  • No accounts, no sign-up, no cookies, no advertising or analytics trackers.
  • We do not sell, rent, or share your data with advertisers or data brokers.
  • The only inputs are public A-share stock codes and screening filters (dividend yield, PE/PB, ROE, market cap). No personal data is required or requested.
  • The server is stateless: each request is independent, with no session and no user profile.
  • Hosting is provided by Apify; platform-level processing is governed by Apify's own privacy policy.

2. Data we process

DataSourceWhy we process itRetention
Stock codes / screener parametersYouTo screen A-shares by dividend/value metricsIn memory for the duration of the request only
Numeric filters (yield, PE/PB, ROE, cap, years)YouTo bound the screening queryRequest duration only
IP address + User-AgentYour requestTransient rate-limiting onlyIn-memory window; not persisted, not logged to disk
Apify API tokenApify gatewayAuthenticates the caller at the platform edgeNot seen or stored by the Service

3. What we do NOT collect

  • No names, email addresses, phone numbers, or other personal identifiers.
  • No trading accounts, brokerage credentials, positions, or portfolio data.
  • No accounts, passwords, or credentials.
  • No conversation history or tool-call content retained beyond the live request.
  • No cookies, analytics, pixels, or advertising trackers.

4. Third parties / data recipients

To answer a query the Service fetches public market data from Eastmoney. Your screener parameters are forwarded to them; no personal data is included.

RecipientPurposeNotes
Eastmoney (data.eastmoney.com, datacenter-web.eastmoney.com)Public A-share fundamentals and dividend dataPublic market endpoints
Apify (hosting)Runs the Standby container and meters usageSubject to Apify's privacy policy

We do not disclose your inputs to any other third party.

5. Hosting and infrastructure

The hosted Service runs on Apify's Standby infrastructure. Apify may process operational metadata (timestamps, IP, billing records) as an independent controller. See https://apify.com/privacy-policy. The Service runs no database and keeps no persistent store of user data.

6. Self-hosted builds

The source is not publicly distributed. Where a self-hosted build exists, its operator is the data controller for anything that deployment processes.

7. Security

Transport is encrypted (TLS) at the Apify edge. All requests require the Apify gateway bearer token. See the repository's SECURITY.md for the threat model and vulnerability reporting.

8. Children's privacy

The Service is a developer tool not directed at children, and we do not knowingly process data from children under 16.

9. Your rights

Because we do not maintain user profiles or store personal data, there is generally no personal data to access, correct, or erase. If you believe we hold data about you, contact us (Section 11) and we will respond within 30 days.

10. Changes to this policy

We may update this policy as the Service evolves. Material changes will be reflected in the "Last updated" date and, where appropriate, in this README.

11. Contact

Privacy questions or requests: contact the operator via this Apify Store listing (Issues tab), or open an issue in the repository if you have access.

隐私政策(简体中文)

产品: China Dividend Value Screener MCP — A 股股息/价值筛选 MCP server 运营方: PanStories 最后更新: 2026-10-09

概要
  • 本服务是只读 MCP server,工具仅读取东方财富免费接口的公开 A 股行情数据, 用于按股息/价值指标选股,不会修改、删除或操作用户的任何系统。
  • 无账号、无注册、无 Cookie、无广告或分析追踪。
  • 我们不会向广告商或数据经纪商出售、出租或共享你的数据。
  • 唯一输入为公开 A 股代码与筛选条件(股息率、PE/PB、ROE、市值),不涉及个人信息。
  • 服务无状态:每次请求独立处理,无会话、无用户画像。
  • 托管由 Apify 提供,平台层处理受 Apify 隐私政策约束。
我们处理的数据
数据来源用途保留
股票代码 / 筛选参数调用方按股息/价值指标筛选 A 股仅请求期间驻留内存
数值筛选(股息率、PE/PB、ROE、市值、年限)调用方限定筛选查询仅请求期间
IP + User-Agent请求仅用于限流内存窗口,不落盘
Apify API tokenApify 网关在平台边缘鉴权本服务不接触、不存储
第三方(仅转发公开数据,不含个人信息)
  • 东方财富(data.eastmoney.com / datacenter-web.eastmoney.com):公开 A 股基本面与分红数据。
  • Apify(托管):运行 Standby 容器并计量。参见 https://apify.com/privacy-policy。
联系方式

通过本 Apify Store 详情页(Issues 标签)联系运营者;如有仓库权限,亦可在仓库提 issue。