ETF Holdings API for AI Agents
4 Providers Tested4 ETF Holdings APIs Tested面向 AI Agent 的 ETF 持仓 API
4 家供应商实测ETF 持仓 API：4 家实测
A constituent-level benchmark of Alpha Vantage, FIU, Gildata, and Wind—measuring full-list coverage, field integrity, invalid-symbol handling, and latency.
对 Alpha Vantage、FIU、Gildata 与 Wind 进行成分股级实测，重点验证完整持仓、字段质量、无效代码处理与响应延迟。
The answer first: which ETF holdings API fits?先说结论：哪一个 ETF 持仓 API 更合适？
For an AI agent that needs constituent-level ETF data, the strongest observed fits were Alpha Vantage for the tested US workflow and Gildata for the tested China workflow. FIU and Wind both returned useful top-holdings slices, but neither returned a complete portfolio in the tested path.
对于需要成分股级 ETF 数据的 AI Agent，本次实测中，Alpha Vantage 最适合美国市场工作流，Gildata 最适合中国市场工作流。FIU 与 Wind 均能返回有用的前十大持仓数据，但在本次测试路径中，两者都没有返回完整组合。
This ETF data API comparison does not assign a blended score. Qualified means the endpoint met the complete response and holdings-integrity requirements in at least 75% of its rounds. Not qualified means it was applicable and tested but missed that requirement. N/A means the product was not applicable to that market workflow.
这份 ETF 数据 API 对比不设置综合评分。“达标”表示接口至少在 75% 的测试轮次中同时满足完整响应和持仓完整性要求；“未达标”表示接口适用于该场景且已完成测试，但没有达到门槛；“N/A”表示产品不适用于相应市场工作流。
| Provider供应商 | US holdings美国持仓 | China holdings中国持仓 | Observed median latency实测中位延迟 |
|---|---|---|---|
| Alpha Vantage Try in QVeris在 QVeris 中试用 | Qualified达标 | N/A | 11.4 s |
| FIU Try in QVeris在 QVeris 中试用 | Not qualified未达标 | N/A | 6.4 s |
| Gildata Try in QVeris在 QVeris 中试用 | N/A | Qualified达标 | 4.8 s |
| Wind Native MCP test · No QVeris ETF connector in this snapshot原生 MCP 测试 · 本次快照中无 QVeris ETF connector | N/A | Not qualified未达标 | 5.9 s |
Access-path note: The access paths are deliberately separated in the evidence and shown together in the charts: Alpha Vantage, FIU, and Gildata were tested through their QVeris connectors, while Wind was tested directly through the official Wind MCP. All four providers appear in both charts, with Wind labeled as the native-MCP track; its row has no QVeris execution link because the dated QVeris ETF Capability snapshot contained no Wind connector.
关于接入路径：证据部分明确区分两种接入方式，但两张图都会展示四家供应商：Alpha Vantage、FIU 与 Gildata 通过各自的 QVeris connector 测试，Wind 则直接通过官方 Wind MCP 测试。图中将 Wind 标注为“原生 MCP”路径；由于测试当时的 QVeris ETF 能力快照中没有 Wind connector，因此 Wind 对应行不提供 QVeris 执行链接。
What AI agents can build with ETF holdings dataAI Agent 可以用 ETF 持仓数据做什么？
ETF holdings data turns a fund ticker into an explainable set of underlying positions. That supports several practical agent workflows:
ETF 持仓数据可以把基金代码展开为一组可解释的底层头寸，从而支持以下实际的 Agent 工作流：
- ETF look-through and portfolio exposure: expand a portfolio of ETFs into underlying securities, then aggregate the effective exposure to each company.ETF 穿透与组合敞口：把 ETF 组合展开为底层证券，再汇总对每家公司的实际敞口。
- Overlap and concentration analysis: find duplicated holdings across funds and show whether a seemingly diversified portfolio is concentrated in the same issuers.重合度与集中度分析：识别不同基金之间的重复持仓，判断一个看似分散的组合是否集中于相同发行人。
- Compliance and restricted-list screening: check whether an ETF creates indirect exposure to a restricted security. This requires a full constituent list and reliable identifiers.合规与受限名单筛查：检查 ETF 是否间接暴露于受限证券。这项任务需要完整成分列表和可靠标识符。
- Rebalancing and substitution research: compare two ETFs by underlying exposure before proposing a replacement or tax-loss candidate. Portfolio actions still require human review and current pricing.再平衡与替代研究：在提出替代 ETF 或税损收割候选方案前，按底层敞口比较两只 ETF；真正执行组合操作仍需人工复核和最新价格。
- Holdings-change monitoring: detect additions, removals, and weight changes between reporting dates. This requires an as-of date and historical snapshots, not just a current response.持仓变动监控：识别不同报告日期之间的新增、剔除和权重变化；这需要持仓日期与历史快照，而不只是当前响应。
- Top-holdings cards and fund briefs: summarize the largest positions for research or client reporting. A clearly labeled top-10 endpoint can support this narrower use even when it cannot support full look-through.前十大持仓卡片与基金简报：为研究或客户报告汇总主要持仓。即使前十大持仓接口无法支持完整穿透，只要明确标注数据范围，仍可用于这类边界清晰的场景。
The data contract depends on the job. Look-through, overlap, and compliance need broad holdings coverage, constituent identifiers, and numeric weights. Change monitoring additionally needs dates and history. Shares, market value, sector, and rank make position cards and exposure explanations richer, but they do not substitute for a full constituent list.
数据契约取决于具体任务。穿透、重合度与合规需要广泛的持仓覆盖、成分股标识符和数值权重；变动监控还需要日期与历史记录。份额、市值、行业和排名能让持仓卡片及敞口解释更丰富，但不能代替完整成分列表。
Benchmark method: 48 observations, one strict contract测试方法：48 次观测，同一套严格标准
Each provider was tested on three valid ETFs and one invalid symbol, with three independent rounds per case. That produced 48 completed observations: 36 QVeris connector invocations and 12 direct Wind MCP invocations. The same completeness gates were applied to both tracks, but results from the native MCP track were not treated as QVeris connector results.
每家供应商都使用 3 只有效 ETF 和 1 个无效代码进行测试，每个用例独立运行 3 轮，共完成 48 次观测：其中 36 次通过 QVeris connector 调用，12 次直接调用 Wind MCP。两条路径采用完全相同的完整性标准，但原生 MCP 路径的结果不会被当作 QVeris connector 的测试结果。
The live run was completed on August 6, 2026.
实测于 2026 年 8 月 6 日完成。
| Market市场 | Provider供应商 | Access path接入路径 | Valid ETF inputs有效 ETF | Negative control无效代码 | Rounds / input每个输入轮次 |
|---|---|---|---|---|---|
| US美国 | Alpha Vantage | QVeris connector | SPY, QQQ, IWM | ZZZZ_INVALID | 3 |
| US美国 | FIU | QVeris connector | SPY.US, QQQ.US, IWM.US | ZZZZ_INVALID.US | 3 |
| China中国 | Gildata | QVeris connector | 510050, 510300, 159915 | 999999 | 3 |
| China中国 | Wind | Official Wind MCP官方 Wind MCP | 510050.SH, 510300.SH, 159915.SZ | 999999.SH | 3 |
A valid holdings response had to clear a case-specific constituent floor based on the fund or tracked index's expected scale, preserve the requested fund identity, provide usable constituent identifiers and numeric weights, and avoid duplicate complete identifiers. Stable fixed-size top-10 responses failed the full-list contract even when their fields were clean. Invalid inputs had to close deterministically as either an explicit empty result or a provider rejection. The US contract allowed up to 2% missing constituent identifiers so that cash or derivative placeholders did not masquerade as duplicate securities; missing identifiers still reduced completeness.
有效的持仓响应必须达到根据基金或跟踪指数预期规模设定的成分股数量门槛，同时保持请求基金身份一致，提供可用的成分股标识符和数值权重，且不能出现完整标识符重复。即使字段本身完整规范，固定只返回前 10 条的响应也无法通过完整列表标准。对于无效输入，接口必须稳定返回明确的空结果，或明确拒绝请求。美国市场测试允许最多 2% 的成分股缺少标识符，以免现金或衍生品占位符被误判为重复证券；但标识符缺失仍会降低完整度。
We report the observed unit and decimal precision of weights instead of using their aggregate as a hard completeness gate. A provider may round every line before returning it; on a high-cardinality fund, those small per-row differences can accumulate materially. Cash, derivatives, and fund-accounting conventions can add further differences. Aggregate weight remains a useful diagnostic when the provider documents its normalization and precision, but it is not evidence by itself that rows are missing.
本次测试记录权重的实际单位与小数精度，但不把权重合计作为判断完整性的硬门槛。供应商可能在返回数据前对每一行进行取整；对于成分数量较多的基金，这些微小的逐行误差可能累积成明显偏差，现金、衍生品和基金核算口径还会造成额外差异。如果供应商明确说明了归一化方法和精度，权重合计仍可作为诊断指标，但它本身不能证明响应缺少记录。
Latency is reported separately and never compensates for incomplete holdings. Prices, redistribution rights, and service-level commitments were not scored.
延迟单独报告，不能弥补持仓数据不完整的问题。价格、数据再分发权和服务等级承诺不在本次评分范围内。
Holdings completeness by ETF按 ETF 查看持仓完整度
The table below reports one stable payload shape observed in all three rounds for each valid input. “Full-list response (tested)” means the endpoint cleared the case-specific record floor and integrity contract in every round; it does not establish the fund administrator's exact point-in-time book. “Top-10 slice” means the request for all holdings still returned only 10 rows. “Reported weight precision” is the maximum decimal precision observed in the response, not a vendor guarantee. dp means decimal places and pp means percentage points.
下表列出每个有效输入在 3 轮测试中保持一致的响应形态。“实测完整持仓”表示接口在每一轮都达到该用例的最低记录数和完整性标准，但不能证明它与基金管理人在某一时点的账簿完全一致。“前十大持仓切片”表示请求全部持仓时，接口仍只返回 10 条记录。“实测权重精度”是本次响应中观测到的最大小数精度，并非供应商承诺；dp 表示小数位，pp 表示百分点。
No complete identifier was duplicated in the observed Alpha Vantage, FIU, or Gildata payloads. Wind returned stable constituent identifiers in all valid runs, but the native MCP response did not expose an as-of date and remained fixed at 10 rows.
在本次观测中，Alpha Vantage、FIU 与 Gildata 的响应均未出现完整标识符重复。Wind 在所有有效测试中也返回了稳定的成分股标识符，但原生 MCP 响应没有提供持仓日期，并且始终固定为 10 条记录。
| Provider供应商 | ETF | Records记录数 | Response scope响应范围 | Reported weight precision实测权重精度 | Additional fields observed观测到的额外字段 | Integrity rounds通过完整性检查的轮次 |
|---|---|---|---|---|---|---|
| Alpha Vantage Try in QVeris在 QVeris 中试用 | SPY | 505 | Full-list response (tested)实测完整持仓 | 4 dp fraction (0.01 pp)小数权重，4 位小数（0.01 个百分点） | name, weight名称、权重 | 3/3 |
| QQQ | 105 | Full-list response (tested)实测完整持仓 | 4 dp fraction (0.01 pp)小数权重，4 位小数（0.01 个百分点） | name, weight名称、权重 | 3/3 | |
| IWM | 1,795 | Full-list response (tested)实测完整持仓 | 4 dp fraction (0.01 pp)小数权重，4 位小数（0.01 个百分点） | name, weight名称、权重 | 3/3 | |
| FIU Try in QVeris在 QVeris 中试用 | SPY.US | 10 | Top-10 slice前十持仓切片 | 5 dp percent百分比权重，5 位小数 | name, weight, as-of date名称、权重、持仓日期 | 0/3 |
| QQQ.US | 10 | Top-10 slice前十持仓切片 | 5 dp percent百分比权重，5 位小数 | name, weight, as-of date名称、权重、持仓日期 | 0/3 | |
| IWM.US | 10 | Top-10 slice前十持仓切片 | 5 dp percent百分比权重，5 位小数 | name, weight, as-of date名称、权重、持仓日期 | 0/3 | |
| Gildata Try in QVeris在 QVeris 中试用 | 510050 | 72 | Full-list response (tested)实测完整持仓 | 4 dp percent百分比权重，4 位小数 | shares, value, as-of date, sector份额、市值、持仓日期、行业 | 3/3 |
| 510300 | 342 | Full-list response (tested)实测完整持仓 | 4 dp percent百分比权重，4 位小数 | shares, value, as-of date, sector份额、市值、持仓日期、行业 | 3/3 | |
| 159915 | 118 | Full-list response (tested)实测完整持仓 | 4 dp percent百分比权重，4 位小数 | shares, value, as-of date, sector份额、市值、持仓日期、行业 | 3/3 | |
| Wind Official native MCP官方原生 MCP | 510050.SH | 10 | Top-10 slice前十持仓切片 | 4 dp percent百分比权重，4 位小数 | name, shares, value, weight, rank名称、份额、市值、权重、排名 | 0/3 |
| 510300.SH | 10 | Top-10 slice前十持仓切片 | 4 dp percent百分比权重，4 位小数 | name, shares, value, weight, rank名称、份额、市值、权重、排名 | 0/3 | |
| 159915.SZ | 10 | Top-10 slice前十持仓切片 | 4 dp percent百分比权重，4 位小数 | name, shares, value, weight, rank名称、份额、市值、权重、排名 | 0/3 |
This matters for agents: an empty portfolio and a rejected request are different states. Both can close a negative control safely when the connector distinguishes them deterministically, so downstream logic does not mistake an error for a real zero-holdings fund.
这一区别对 Agent 很重要：空组合与请求被拒绝是两种不同状态。只要 connector 能稳定区分两者，无效输入测试就能安全结束，避免下游逻辑把错误状态误认为基金确实没有持仓。
| Provider供应商 | Invalid-input behavior无效输入行为 | Result结果 |
|---|---|---|
| Alpha Vantage | Deterministic request rejection with parameter guidance稳定地拒绝请求，并返回参数提示 | 3/3 passed通过 |
| FIU | Empty constituent result返回空成分股结果 | 3/3 passed通过 |
| Gildata | Empty holdings result返回空持仓结果 | 3/3 passed通过 |
| Wind | Explicit “no data found” result from the native MCP原生 MCP 明确返回“未找到数据”结果 | 3/3 passed通过 |
Provider-by-provider findings逐家供应商实测发现
The live Alpha Vantage ETF profile calls returned broad constituent lists, weights, and descriptions. SPY, QQQ, and IWM all cleared the full-list contract in every round; IWM consistently returned 1,795 rows. Several cash or derivative rows used placeholder identifiers, which reduced identifier coverage without being misclassified as duplicate securities.
Alpha Vantage ETF Profile 的实测调用返回了覆盖范围较广的成分股列表、权重和说明。SPY、QQQ 与 IWM 在每一轮都通过完整列表标准，其中 IWM 稳定返回 1,795 条记录。部分现金或衍生品行使用占位标识符，这会降低标识符覆盖率，但不会被误判为重复证券。
Alpha Vantage reported weights as fractions with at most four decimal places, equivalent to 0.01 percentage-point granularity. In IWM, 221 rows were reported with zero weight at that precision. That makes the aggregate unsuitable as a hard row-completeness test. The response also lacked an as-of date, so the benchmark supports “full-list response” but does not prove exact point-in-time completeness against the fund administrator's book.
Alpha Vantage 以小数形式报告权重，最多保留四位小数，相当于 0.01 个百分点的精度。IWM 中有 221 行在该精度下显示为零，因此权重合计不适合作为判断行级完整性的硬门槛。响应也没有提供持仓日期，所以本次测试只能支持“返回完整列表”的结论，不能证明它与基金管理人某一时点的账簿完全一致。
The Alpha Vantage ETF profile documentation is the official contract reference. Editorially, this was the strongest observed fit in this test for a US agent that needs broad portfolio holdings in one response. The exact connector is available on the Alpha Vantage provider page in QVeris.
Alpha Vantage ETF Profile 官方文档是该接口契约的权威参考。就本次实测而言，它最适合需要一次获取较完整美国 ETF 组合持仓的 Agent。对应 connector 可从 QVeris 的 Alpha Vantage 供应商页面直接查看。
The live FIU payloads were internally tidy: identifiers, names, weights, parent symbols, and publication dates were present for all returned rows. The limiting factor was scope. Each valid input returned exactly 10 constituents, far below the case-specific floors for SPY, QQQ, and IWM.
FIU 的实测响应字段完整且规范：所有返回行都包含标识符、名称、权重、所属基金代码和发布日期。真正的限制在于数据范围——每个有效输入都固定返回 10 只成分股，远低于 SPY、QQQ 和 IWM 各自的完整持仓门槛。
The FIU MCP server is the declared official source. Editorially, the observed endpoint may suit top-holdings summaries, but it should not be presented to an agent as a full holdings API without pagination or a broader constituent route. The tested connector is listed on the FIU provider page in QVeris.
FIU MCP Server 是其声明的官方来源。就本次观测而言，该接口可用于生成前十大持仓摘要；但在没有分页或更完整成分股接口的情况下，不应把它作为完整持仓 API 提供给 Agent。实测 connector 可在 QVeris 的 FIU 供应商页面查看。
The live Gildata responses passed all 12 applicable observations. Besides fund and constituent identities, names, and weights, every returned row included shares, market value, an as-of date, and sector information. The three portfolios returned 72, 342, and 118 records, clearing their case-specific floors rather than repeating an artificial fixed-size slice.
Gildata 的实测响应通过了全部 12 次适用观测。除了基金与成分股标识、名称和权重外，每条返回记录还包含份额、市值、持仓日期与行业信息。三只 ETF 分别返回 72、342 和 118 条记录，都达到各自用例的最低数量标准，而不是固定返回相同条数的数据切片。
The Gildata official site describes its financial-data offering. Editorially, Gildata was the strongest observed China fit in this test when an agent needs both composition and position-level depth. The connected entry is available on the Gildata provider page in QVeris.
Gildata 官网介绍了其金融数据服务。就本次实测而言，当 Agent 同时需要组合构成和详细的持仓级数据时，Gildata 是中国市场适配度最高的方案。对应入口可在 QVeris 的 Gildata 供应商页面查看。
The official Wind MCP get_fund_holdings tool returned exactly 10 rows for each valid ETF in all nine positive observations, even though every request explicitly asked for all holdings rather than only the top 10. The rows were information-rich: constituent codes and names, shares, market values, weights, and ranks were all present. The fixed 10-row shape, requested-all mismatch, and failure to reach the case-specific record floors identify these as top-holdings slices. No as-of date was included in the response.
官方 Wind MCP 的 get_fund_holdings 工具在 9 次有效观测中，每次都只返回 10 条记录，尽管所有请求都明确要求返回全部持仓，而不是仅返回前十。返回字段较为丰富，包括成分股代码与名称、份额、市值、权重和排名。但响应始终固定为 10 条、与“请求全部持仓”的要求不一致，而且未达到各用例的最低记录数，因此只能视为前十大持仓切片。响应中也没有提供持仓日期。
The Wind AI financial data marketplace is the official access point, and Wind publishes the MCP integration in its official Wind skills repository. Editorially, this tool fits top-holdings summaries and position cards, but it should not be used as a complete ETF allocation feed without a separate full-holdings route.
Wind AI 金融数据市场是官方接入入口，Wind 也在官方 Wind Skills 仓库发布了 MCP 集成。就本次实测而言，该工具适合生成前十大持仓摘要和仓位卡片；如果没有其他能够返回完整持仓的接口，就不应把它作为完整的 ETF 配置数据源。
How to choose an ETF portfolio API for an agent如何为 Agent 选择 ETF 组合数据 API？
- For US look-through, overlap, or compliance screening, Alpha Vantage was the tested full-list fit; add your own freshness control because the response did not expose an as-of date.如果要做美国市场的持仓穿透、重合度分析或合规筛查，Alpha Vantage 是本次实测中的完整持仓方案；由于响应没有提供持仓日期，仍需自行增加数据时效校验。
- For a top-holdings card, FIU's compact response may be useful if the product explicitly labels it as partial.如果只需制作前十大持仓卡片，可以使用 FIU 的精简响应，但产品必须明确标注这只是部分持仓。
- For mainland China composition plus shares, value, date, and sector fields, Gildata had the strongest observed contract fit.如果需要中国内地 ETF 的组合构成，同时还需要份额、市值、日期和行业字段，Gildata 在本次观测中与接口字段要求的匹配度最高。
- For a China top-10 holdings card that benefits from shares, market value, weight, and rank, Wind's native MCP is useful; do not treat the tested response as the full portfolio.如果要做包含份额、市值、权重和排名的中国市场前十大持仓卡片，Wind 原生 MCP 很有用；但不能把本次实测响应当成完整组合。
For implementation patterns beyond this comparison, see Portfolio monitoring agent and Financial data for AI agents.
如果需要本次对比以外的实现方法，可继续阅读组合监控 Agent和面向 AI Agent 的金融数据能力。
Limitations and retesting rules测试边界与复测规则
This edition covers constituent holdings for six named funds across US and China workflows. It does not establish complete exchange coverage or evaluate NAV, AUM, flows, corporate-action handling, survivorship bias, point-in-time corrections, pricing, redistribution rights, or contractual SLAs. The Alpha Vantage, FIU, and Gildata observations describe QVeris-connected behavior on August 6, 2026; the Wind observations describe the official native MCP on the same date. Provider plans, connectors, or later responses may differ.
本版测试覆盖美国与中国工作流中的 6 只指定基金，仅讨论成分持仓。不能据此判断供应商是否覆盖全部交易所；本次测试也没有评估 NAV、AUM、资金流、公司行动处理、幸存者偏差、历史时点数据修订、定价、再分发权限或合同 SLA。Alpha Vantage、FIU 与 Gildata 的观测结果反映了 2026 年 8 月 6 日通过 QVeris connector 的实际表现；Wind 的观测结果则来自同日的官方原生 MCP。供应商套餐、connector 或后续响应都可能发生变化。
Suppliers can submit a correction or propose a reproducible test case. Inclusion and placement cannot be purchased. Browse the QVeris Provider Hub or browse QVeris tools.
供应商可以提交更正信息或提出可复现的测试用例，但不能付费购买入选或排名。你也可以浏览 QVeris 供应商目录或 QVeris 工具目录。
FAQ
Which tested API returned complete US ETF portfolios?哪个实测接口返回了完整的美国 ETF 组合？
Alpha Vantage qualified. SPY, QQQ, and IWM passed all three full-list rounds, and the invalid control closed deterministically. IWM returned 1,795 rows; its four-decimal fractional weights were treated as reported precision, not as a hard aggregate completeness test.
Alpha Vantage 达标。SPY、QQQ 与 IWM 的 3 轮完整列表测试全部通过，无效代码测试也能稳定结束。IWM 返回 1,795 条记录；四位小数的权重被视为供应商报告精度，而不是用权重合计判断完整性的硬门槛。
Why did FIU not qualify if all fields were present?FIU 字段齐全，为什么仍未达标？
Field presence is not portfolio completeness. FIU returned exactly 10 rows per valid ETF, below the case-specific record floors needed for a full holdings workflow.
字段齐全不等于组合完整。FIU 对每只有效 ETF 固定返回 10 条，低于完整持仓工作流中各用例所需的最低记录数。
Which provider exposed the deepest China fields?哪家供应商提供的中国 ETF 持仓字段最丰富？
Gildata returned weights, shares, market values, as-of dates, and sectors for all tested rows. Wind returned shares, market values, weights, and ranks for the top 10, but no response-level as-of date or complete portfolio.
Gildata 为全部实测记录提供了权重、份额、市值、持仓日期和行业字段。Wind 为前十大持仓提供了份额、市值、权重和排名，但没有提供响应级持仓日期，也没有返回完整组合。
Why is there no overall numeric score?为什么不提供综合量化评分？
A single score would hide whether an endpoint returns a complete portfolio or only a partial list. The qualification states and per-ETF evidence keep those trade-offs visible.
单一分数会掩盖接口返回的是完整组合还是部分列表。分别展示达标状态和逐 ETF 证据，可以让这些关键差异保持清晰。