Connect Gemini to a read-only market-data source through MCP or a composable CLI—then require every result to carry a symbol, price, currency, market session, source, and timestamp. 通过 MCP 或可组合的 CLI，把 Gemini 接入只读行情数据源，并要求每次结果都包含股票代码、价格、币种、交易时段、来源和时间戳。
Use this guide to choose an integration path, define freshness correctly, implement the tool contract, and validate what Gemini says when market conditions are messy.本指南帮助你选择接入方式、正确定义数据新鲜度、实现工具契约，并验证 Gemini 在复杂市场状态下给出的回答。
The right path depends on where the conversation lives and how much control the workflow needs. 选择取决于对话发生在哪里，以及工作流需要多少控制权。
Best when Gemini should discover and call structured market capabilities directly. Configure an MCP server, keep credentials outside prompts and repositories, inspect the chosen tool, and allow only the read-only quote operations the task needs. 适合让 Gemini 直接发现并调用结构化行情能力。配置 MCP Server，把密钥留在提示词和代码仓库之外，先检查所选工具，并只开放任务所需的只读行情操作。
Read the QVeris MCP server setup 阅读 QVeris MCP Server 配置说明Best when a repository needs a testable, versioned adapter. Give Gemini one command that validates symbols, calls the provider, emits stable JSON, and returns non-zero exit codes for timeouts, limits, or ambiguous instruments. 适合在代码仓库中维护可测试、可版本化的适配器。为 Gemini 提供一条命令：校验股票代码、调用数据源、输出稳定 JSON，并在超时、限流或标的歧义时返回非零退出码。
Open the official Gemini CLI MCP setup 打开 Gemini CLI 官方 MCP 配置指南Run the official Gemini command below with your own API key. It registers the read-only capability layer in Gemini without putting the key in the repository. The Gemini CLI and IDE extension store this MCP entry in your local Gemini configuration.使用你自己的 API key 运行下面的 Gemini 官方配置命令。它会把只读能力层注册到 Gemini，同时避免把密钥写进代码仓库。Gemini CLI 与 IDE 扩展会把该 MCP 条目保存在本机 Gemini 配置中。
gemini mcp add qveris --env QVERIS_API_KEY=your-api-key-here -- npx -y @qverisai/mcp
gemini mcp listAfter QVeris appears as enabled, ask Gemini to discover a capability before executing anything. Inspect the selected capability to confirm its provider, input schema, freshness, entitlement, and cost. Only then call it with parameters that match the inspected schema.确认 QVeris 已启用后，先让 Gemini 发现能力，不要直接执行。检查所选能力的数据源、输入 schema、新鲜度、授权和成本，再按照检查得到的 schema 发起调用。
qveris discover "real-time stock quote with source and timestamp"
qveris inspect <tool-id>
qveris call <tool-id> --params '<validated-json-from-the-inspected-schema>'Verification prompt: “Use QVeris to find a read-only current quote capability for AAPL. Inspect it first. Return the exchange, currency, market session, source, delay status, and provider timestamp. Do not place a trade.”验证提示词：“使用 QVeris 查找查询 AAPL 当前行情的只读能力，先检查再调用。返回交易所、币种、交易时段、来源、延迟状态和数据源时间戳，不要执行交易。”
See the official Gemini CLI MCP setup and troubleshooting steps查看 Gemini CLI 官方 MCP 配置与故障排查步骤
Adding real-time stock market data to Gemini is not a direct model-to-exchange connection. A production design separates language reasoning from deterministic validation, provider access, and evidence formatting.为 Gemini 接入实时股票市场数据，并不是让模型直接连接交易所。生产架构需要把语言推理与确定性的参数校验、数据源访问和证据格式化分开。
Gemini identifies whether the user wants a current quote, a historical comparison, market status, or a broader research task. It should ask for clarification when a symbol, exchange, or time horizon is ambiguous.Gemini 判断用户要查询当前行情、历史对比、市场状态，还是更完整的研究任务。若股票代码、交易所或时间范围有歧义，应先追问确认。
A narrow schema constrains symbols, exchanges, fields, and permitted operations. This layer rejects malformed input, applies user and workspace policy, and keeps read-only retrieval separate from trading.边界清晰的 schema 约束股票代码、交易所、返回字段和允许操作。本层拒绝异常输入、执行用户与工作区策略，并把只读查询和交易严格分离。
Server-side code owns credentials, provider selection, entitlement checks, retries, normalization, and caching. If providers use different names for sessions or timestamps, the adapter maps them into one stable response.服务端代码负责凭证、数据源选择、授权检查、重试、标准化和缓存。若不同数据源对交易时段或时间戳使用不同字段，适配层应统一为稳定结构。
The final response combines the quote with its currency, as-of time, session, delay status, and source. The model may explain the result, but it must not silently replace missing fields or present stale data as current.最终回答把行情与币种、数据时间、交易时段、延迟状态和来源一起呈现。模型可以解释结果，但不能悄悄补齐缺失字段，也不能把旧数据说成当前数据。
Let the model decide when a quote is needed; let application code decide whether and how the request is safe to execute. The returned JSON—not model memory—is the source of truth for the price.由模型判断何时需要行情；由应用代码判断请求是否可以执行以及如何执行。价格的事实来源是工具返回的 JSON，而不是模型记忆。
A price without timing context can be confidently wrong. Require enough metadata to explain what the number represents. 缺少时间上下文的价格很容易造成误导。应强制返回足够的元数据，说明这个数字究竟代表什么。
| Field 字段 | Why it matters 为什么重要 | Example shape 示例形式 |
|---|---|---|
| symbol, exchange | Disambiguates tickers across venues. 避免不同交易所出现同名代码时混淆。 | NASDAQ:AAPL |
| price, currency | Prevents an unlabeled number from being treated as USD. 防止把没有币种标识的数字默认当成美元。 | number + ISO currency |
| as_of | Shows when the provider observed or published the value. 说明数据源观察或发布该值的时间。 | ISO 8601 timestamp |
| session | Separates regular, pre-market, after-hours, and closed states. 区分正常交易、盘前、盘后和休市。 | regular / pre / post / closed |
| delay, source | Makes licensing and freshness limits visible. 把授权范围和新鲜度限制明确展示出来。 | delay seconds + provider |
The best stock market API for Gemini is the one whose coverage, entitlement, freshness, and response metadata match the questions your users ask. A cheap quote endpoint may be sufficient for casual lookup but unsuitable for alerts, compliance-sensitive research, or redistribution.最适合 Gemini 的股票行情 API，应在市场覆盖、授权、新鲜度和响应元数据上匹配用户问题。低成本行情端点可能足够应付普通查询，却未必适合提醒、合规敏感研究或数据再分发。
| Criterion评估维度 | Questions to verify需要核实的问题 | Failure if ignored忽略后的风险 |
|---|---|---|
| Market coverage市场覆盖 | Which exchanges, instruments, countries, and asset classes are included?覆盖哪些交易所、证券类型、国家和资产类别？ | The same ticker resolves to the wrong venue or no result.同名代码被解析到错误市场，或完全没有结果。 |
| Freshness新鲜度 | Is the value real-time, delayed, indicative, snapshot, or previous close for this entitlement?在当前授权下，返回的是实时、延迟、指示性、快照还是前收盘价？ | Gemini labels an old value as a live stock price.Gemini 把旧价格误称为实时股价。 |
| Session detail交易时段 | Can the API distinguish regular, pre-market, after-hours, halted, and closed?API 能否区分正常交易、盘前、盘后、停牌和休市？ | A post-market value is compared with a regular-session close without explanation.盘后价格与正常收盘价被直接比较，却没有说明口径。 |
| Licensing数据授权 | May values be displayed to end users, cached, logged, or redistributed?数据能否向终端用户展示、缓存、记录或再分发？ | A technically working integration violates display terms.技术上可用的集成却违反了展示条款。 |
| Operational fit运维适配 | What are the rate limits, timeout behavior, status page, and support path?限流、超时行为、状态页和支持渠道分别是什么？ | Peak traffic becomes unexplained missing or repeated quotes.高峰流量导致行情缺失或重复，却无法定位原因。 |
Do not copy a universal “real-time” label from a provider homepage. Verify the specific exchange, endpoint, account tier, and intended display use in the provider’s official documentation and agreement.不要直接照搬数据源首页上的“实时”宣传语。应在官方文档和协议中核实具体交易所、端点、账户级别及预期展示用途。
Start with quote lookup, daily change, session status, and a timestamp. Treat charts, news, filings, screening, and trading as separate capabilities with separate permissions. 先支持行情查询、日内涨跌、交易时段状态和时间戳。图表、新闻、公告、选股与交易应作为权限独立的能力处理。
Check exchange coverage, entitlement, delay, redistribution terms, uptime, rate limits, and whether timestamps and session status are returned. “Real-time” must match your plan and market. 核对交易所覆盖、数据授权、延迟、再分发条款、可用性、限流，以及是否返回时间戳和交易状态。“实时”的定义必须与你购买的套餐和市场一致。
Accept a symbol or query plus an optional exchange. Return a stable JSON object. Keep provider keys server-side and never ask the model to assemble secret-bearing URLs. 输入股票代码或查询词，并可选指定交易所；输出稳定的 JSON。数据源密钥只保留在服务端，不要让模型拼接带密钥的 URL。
Require the answer to state the as-of time, currency, session, and source. If data is stale, missing, or ambiguous, the assistant should say so rather than infer a current price. 要求回答注明数据时间、币种、交易时段与来源。若数据过期、缺失或代码有歧义，助手应明确说明，不得推测当前价格。
Test weekends, holidays, halts, pre-market and after-hours, invalid symbols, duplicate ticker names, provider timeouts, rate limits, and split-adjusted historical values. 测试周末、节假日、停牌、盘前盘后、无效代码、重名代码、数据源超时、限流，以及复权历史数据。
Log request ID, selected capability, provider, latency, status, and data timestamp—without leaking keys. Cache only for a duration compatible with the promised freshness. 记录请求 ID、所选能力、数据源、延迟、状态和数据时间戳，但不要泄露密钥。缓存时长必须符合对用户承诺的新鲜度。
This illustrative tool definition is intentionally provider-neutral. Your application—not the model—owns authentication, symbol normalization, retries, entitlements, and the final provider request. 下面的工具定义是与数据源无关的示例。认证、代码标准化、重试、授权判断和最终的数据源请求，都应由你的应用而不是模型负责。
Separate factual market retrieval from recommendations, trading, suitability assessment, and regulated activities. 应把客观行情查询与推荐、交易、适当性评估及受监管活动明确分开。
{
"type": "function",
"name": "get_stock_quote",
"description": "Return a current market quote with freshness metadata",
"parameters": {
"type": "object",
"properties": {
"symbol": { "type": "string" },
"exchange": { "type": "string" }
},
"required": ["symbol"],
"additionalProperties": false
}
}
// Server returns:
{
"symbol": "...",
"price": 0,
"currency": "...",
"session": "...",
"as_of": "...",
"delay_seconds": 0,
"source": "..."
}
A good tool call can still produce a misleading answer if the model hides timing and scope. Give the assistant a compact output policy that is easy to test.即使工具调用正确，如果模型隐藏时间和口径，最终回答仍可能误导。应给助手一套简洁、可测试的输出规则。
Example Company (EXCH:XYZ) is shown at 123.45 USD. The quote is marked regular session and was observed at 14:32:08 America/New_York. Source: Example Provider. Data status: real-time under the configured entitlement.示例公司（EXCH:XYZ）当前显示为 123.45 USD。该行情标记为正常交易时段，数据时间为 14:32:08 America/New_York。来源：示例数据源。数据状态：在当前授权下为实时。
This is a formatting example, not an observed market quote. Production values must come from the tool result.这只是格式示例，不是真实观测行情。生产环境中的数值必须来自工具返回结果。
“For any request containing current, live, now, today, latest, pre-market, or after-hours price intent, call the quote tool before answering. Never recall a price from conversation history. If the result lacks an as-of timestamp, source, or session status, state that the quote cannot be verified as current.”“凡请求包含当前、实时、现在、今天、最新、盘前或盘后价格意图，回答前必须调用行情工具。不得从对话历史中回忆价格。如果结果缺少数据时间、来源或交易时段状态，应说明无法验证该行情是否为当前数据。”
QVeris can help an agent discover market-data capabilities, inspect their inputs and outputs, and call the selected tool through a unified interface. This is useful when one workflow may later expand from quotes to filings, news, indicators, or alerts. QVeris 可帮助 Agent 发现市场数据能力、检查输入输出，并通过统一接口调用选定工具。当工作流可能从行情扩展到公告、新闻、指标或提醒时，这种方式尤其有用。
Use the live stock page to inspect quote freshness, charts, and market-session context. 通过实时股票页检查行情新鲜度、图表和交易时段上下文。
Open QVeris Real-Time Market Viewer 打开 QVeris 实时市场查看器For code-first workflows, discover and inspect stock capabilities through the QVeris CLI or documentation. 对于代码优先的工作流，可通过 QVeris CLI 或文档发现并检查股票能力。
Read the QVeris CLI guide 阅读 QVeris CLI 指南See how a market-data MCP entry point can cover quotes, movers, news, macro signals, and historical data. 了解市场数据 MCP 入口如何覆盖行情、异动、新闻、宏观信号和历史数据。
Read the market data MCP guide 阅读市场数据 MCP 指南A reliable Gemini stock-data workflow makes these trade-offs explicit. Optimizing only for the fastest model response can increase quote staleness, provider cost, or operational risk.可靠的 Gemini 股票数据工作流必须明确这些取舍。只追求模型最快回答，可能增加行情陈旧、数据成本或运维风险。
Set cache policy by user intent and feed rights. A repeated educational lookup may tolerate a short cache; a market alert may require a new request. Return both the provider timestamp and any cache age so the assistant can describe the result honestly.根据用户意图和数据授权设置缓存策略。重复的教学查询可能允许短时缓存，市场提醒则可能必须重新请求。应同时返回数据源时间戳和缓存年龄，让助手如实描述结果。
Use a bounded timeout and at most a controlled retry for transient failures. If a fallback provider has different coverage or delay, expose that substitution in the response instead of making the sources appear equivalent.设置明确超时，并只对暂时性故障进行有限重试。若备用数据源在覆盖或延迟上不同，应在回答中说明替换，不能让不同来源看起来完全等价。
Prevent one prompt from expanding into hundreds of symbols. Cap list size, separate batch endpoints, deduplicate repeated symbols, and monitor provider calls per conversation. Ask for confirmation before expensive scans.避免一个提示词扩展成数百个股票查询。限制列表规模、为批量查询使用独立端点、去重重复代码，并监控每次对话的数据源调用数。高成本扫描前应要求确认。
Store market-data keys in a server-side secret manager or managed connection. Scope credentials to read-only data where possible, rotate them, redact request headers from logs, and never return provider secrets to Gemini.把市场数据密钥保存在服务端密钥管理器或托管连接中。尽可能使用只读权限，定期轮换，在日志中隐藏请求头，并且绝不能把数据源密钥返回给 Gemini。
A quote symbol is often low sensitivity, but watchlists, positions, account identifiers, and strategy prompts may not be. Minimize what is sent to the provider and document what external service receives.单个股票代码通常敏感度较低，但自选列表、持仓、账户标识和策略提示词可能并非如此。应减少发送给数据源的内容，并明确哪些外部服务会接收数据。
Keep the quote tool read-only. If trading is ever added, use a separate tool, stronger authentication, explicit confirmation, idempotency, position and risk checks, and a complete audit trail.行情工具应保持只读。如果未来增加交易能力，应使用独立工具、更强认证、明确确认、幂等控制、持仓与风险检查，以及完整审计记录。
| Symptom现象 | Likely cause可能原因 | What to fix修复方法 |
|---|---|---|
| Gemini gives a price without calling the toolGemini 未调用工具就给出价格 | Tool description or assistant instructions do not identify freshness intent.工具描述或助手指令没有明确识别实时意图。 | Add trigger terms and require a tool call for current-price questions; test paraphrases, not only one prompt.补充触发词，并要求当前价格问题必须调用工具；测试多种表达，而不只是一个提示词。 |
| The quote is correct but labeled “live” incorrectly价格正确，但被错误标成“实时” | The adapter discarded provider delay or entitlement metadata.适配层丢弃了数据源延迟或授权信息。 | Preserve data status and make the model repeat it; never infer real-time status from a recent timestamp alone.保留数据状态并要求模型复述；不能仅凭时间戳较新就推断为实时。 |
| A symbol returns the wrong company股票代码返回了错误公司 | Ticker resolution ignored exchange or country context.代码解析忽略了交易所或国家上下文。 | Return candidate matches and ask the user to disambiguate before the quote call.先返回候选结果，在调用行情前让用户确认。 |
| Prices change between two turns with no explanation两轮对话价格变化，却没有解释 | One answer came from cache or a different session/feed.其中一次来自缓存，或使用了不同交易时段/数据流。 | Display as-of time, cache age, session, and source on every answer.每次回答都显示数据时间、缓存年龄、交易时段与来源。 |
| The assistant retries until rate-limited助手持续重试直至触发限流 | The tool loop lacks terminal error states and retry limits.工具循环缺少终止错误状态和重试上限。 | Classify errors as retryable or terminal, cap attempts, and return a user-readable failure reason.把错误分为可重试和终止型，限制尝试次数，并返回用户可理解的失败原因。 |
A useful evaluation set measures more than whether an API returned HTTP 200. Test tool selection, argument accuracy, evidence preservation, language quality, and safe failure behavior.有效的评测不应只检查 API 是否返回 HTTP 200，还要测试工具选择、参数准确性、证据保留、语言质量和安全失败行为。
Do not publish the integration until every supported market has an explicit freshness definition, every quote includes traceable metadata, and every tested provider failure ends with an honest user-facing response. Re-run the set whenever the provider, schema, tool description, model, or caching policy changes.在每个支持市场都有明确新鲜度定义、每条行情都具备可追踪元数据、每种已测试的数据源故障都能得到诚实的用户回答之前，不应上线。更换数据源、schema、工具描述、模型或缓存策略后，都要重新执行评测。
Do not assume it does. For a repeatable application, connect a trusted market-data source and retrieve the quote at answer time. Display the source, market session, and timestamp. 不要作此假设。若要构建可重复运行的应用，应连接可信行情数据源，在回答时即时查询，并展示来源、交易时段和时间戳。
Yes. Connect an approved MCP server or give Gemini a narrow CLI that calls the provider and returns structured JSON. Keep credentials out of the repository, make the quote path read-only, and test failure behavior before relying on it. 可以。可连接获准使用的 MCP Server，或为 Gemini 提供调用数据源并返回结构化 JSON 的专用 CLI。密钥不要写入仓库，行情链路应保持只读，并在正式使用前测试失败行为。
The exact meaning depends on the exchange, feed, and entitlement. A provider may return a current value for one venue but delayed or indicative data for another. Preserve the provider timestamp and delay metadata instead of applying one label globally. 具体含义取决于交易所、数据流和授权。同一数据源可能对某个市场提供实时值，对另一个市场只提供延迟或指示性数据。因此应保留来源时间戳与延迟信息，不能统一笼统标注。
Usually no. Keep read-only market retrieval separate from order entry. Trading requires stronger identity, authorization, confirmations, risk controls, audit trails, and regulatory review. 通常不应。只读行情查询应与下单分离。交易还需要更严格的身份认证、授权、确认、风险控制、审计记录和监管评估。
Return the latest available value with its timestamp and session status, and state whether it is the previous close, an after-hours quote, or another provider-defined value. 返回最新可用值，同时给出时间戳和交易时段状态，并说明它是前收盘价、盘后行情，还是数据源定义的其他数值。
Google product behavior and market-data entitlements can change. Check the current official documentation before publishing or modifying a production integration.Google 产品行为和市场数据授权都可能变化。发布或修改生产集成前，应核对最新官方文档。
Start with one symbol, verify its source and freshness, and inspect the structured result before adding news, filings, indicators, or alerts. 从一个股票代码开始，核对来源与新鲜度并检查结构化结果，然后再添加新闻、公告、指标或提醒。
Give Gemini current quotes by connecting a licensed market-data source through Gemini API function calling or a Gemini CLI MCP server. Gemini chooses the tool; your application retrieves and validates the data.通过 Gemini API 函数调用或 Gemini CLI MCP Server 连接具备授权的行情源，让 Gemini 获取当前报价。Gemini 负责选择工具，应用负责查询与校验数据。
These routes solve different deployment problems. Do not combine them unless the same product genuinely supports both an API application and an interactive developer workflow.两条路径对应不同的部署场景。只有产品同时包含 API 应用和交互式开发工作流时，才需要同时采用。
Declare a narrow function such as
get_stock_quote. Gemini returns structured
arguments; your backend validates the symbol, calls the
provider, and sends the function result back for a final
answer. Google’s documentation explicitly places function
execution in application code, not inside the model.声明一个边界清晰的函数，例如
get_stock_quote。Gemini
返回结构化参数；后端校验股票代码、调用数据源，再把函数结果发回模型生成最终回答。Google
官方流程明确指出，函数由应用代码执行，而不是由模型直接执行。
Register a market-data MCP server in Gemini CLI. The CLI discovers its tools and exposes their schemas to Gemini. Keep the server untrusted until you have inspected it, pass secrets through environment variables, and include only read-only tools needed for research.在 Gemini CLI 中注册市场数据 MCP Server。CLI 会发现工具并把 Schema 提供给 Gemini。检查完成前不要授予信任；密钥通过环境变量传入，并且只开放研究需要的只读工具。
Follow the official Gemini CLI MCP tutorial按照 Gemini CLI 官方 MCP 教程配置The critical Gemini-specific detail is the function-call round trip: the first model response asks for a function; the second receives the verified result. Never let the model invent the execution result.Gemini 接入的关键是函数调用往返：第一轮由模型请求函数，第二轮接收经过校验的结果。绝不能让模型自行编造执行结果。
{
"type": "function",
"name": "get_stock_quote",
"description": "Get a current read-only stock quote",
"parameters": {
"type": "object",
"properties": {
"symbol": {"type": "string"},
"exchange": {"type": "string"}
},
"required": ["symbol"]
}
}Gemini CLI stores MCP definitions in its settings and can verify connections with its MCP commands. For QVeris, start by reading the current server documentation rather than copying an unverified package command from an older page.Gemini CLI 会在设置中保存 MCP 定义，并可通过 MCP 命令验证连接。接入 QVeris 时，应先查看当前 Server 文档，不要从旧页面复制未经验证的软件包命令。
gemini mcp list or
/mcp list and confirm it connects.运行 gemini mcp list 或
/mcp list，确认连接成功。
“Find a read-only quote tool for AAPL. Before calling it, show the selected tool and required parameters. Return exchange, currency, session, source, delay status, and provider timestamp. Do not trade.”“查找可查询 AAPL 当前行情的只读工具。调用前先显示所选工具和必需参数。返回交易所、币种、交易时段、来源、延迟状态和数据源时间戳。不要执行交易。”
Check the current QVeris MCP server documentation查看最新 QVeris MCP Server 文档
A bare price is not enough for a current-market answer. Normalize the provider payload before it enters the model context.只有价格不足以支撑当前行情回答。数据进入模型上下文前，应先规范化数据源载荷。
| Field字段 | Why Gemini needs itGemini 为什么需要 | Failure to prevent需要避免的问题 |
|---|---|---|
symbol + exchange |
Identifies the instrument unambiguously.明确标的身份。 | Answering for the wrong venue.回答了错误市场的标的。 |
price + currency |
Preserves value and unit together.同时保留数值与单位。 | Assuming every quote is USD.误把所有价格当成美元。 |
as_of + timezone |
States when the provider observed the quote.说明数据源观测行情的时间。 | Presenting an old value as current.把旧数据说成当前数据。 |
session |
Separates regular, pre-market, after-hours, and closed.区分正常交易、盘前、盘后与休市。 | Comparing incompatible session values.比较不同时段的不可比价格。 |
source + delay_status |
Makes provenance and freshness explicit.明确来源与新鲜度。 | Calling delayed data “real time.”把延迟行情称为实时行情。 |
Confirm the exchange, feed, account plan, endpoint, and display rights. Technical API access alone does not prove that the returned quote is real-time or redistributable. Alpaca’s official documentation is one example of a provider explaining feed and subscription differences.需要核查交易所、Feed、账户套餐、端点与展示权利。能够调用 API，并不代表返回数据一定实时或允许再分发。Alpaca 官方文档就是数据源说明 Feed 与订阅差异的一个例子。 Review an official real-time feed explanation查看官方实时 Feed 说明
“Real-time stock market data for Gemini” can describe three very different workloads. Choosing the wrong one either wastes provider calls or gives the model more data than it can use responsibly.“Gemini 实时股票市场数据”可能对应三类完全不同的工作负载。选错模式，要么浪费数据源调用，要么向模型塞入超出其可靠处理范围的数据。
Use a REST-style quote function when a user asks for the current price, day range, volume, or market status of a small number of symbols. This is the default pattern for Gemini chat experiences because the tool result is compact, auditable, and easy to associate with one answer.当用户查询少量股票的当前价格、日内区间、成交量或市场状态时，使用 REST 风格的报价函数。这是 Gemini 对话体验的默认模式，因为工具结果紧凑、可审计，也容易与单次回答对应。
For opening, closing, or portfolio briefs, let a scheduler collect an approved symbol set and normalize the results before Gemini sees them. The model should summarize a completed batch—not decide silently to expand one prompt into hundreds of billable requests.对于开盘、收盘或组合简报，应由调度器先查询批准的股票集合并规范化结果，再交给 Gemini。模型负责总结已完成的批次，而不是悄悄把一个提示词扩展为数百次付费请求。
For WebSocket feeds, a deterministic service should maintain state, calculate thresholds, deduplicate events, and emit only meaningful alerts. Call Gemini after a trigger to explain context or draft a notification. An LLM is not the place to calculate every price change in a high-frequency stream.对于 WebSocket Feed，应由确定性服务维护状态、计算阈值、去重事件，只输出有意义的告警。触发后再调用 Gemini 解释背景或撰写通知。高频数据流中的每次价格变化不应交给 LLM 计算。
A reliable integration is a stateful application workflow, not a clever prompt. Each boundary has a different owner and a different failure response.可靠接入是有状态的应用工作流，而不是一段巧妙的 Prompt。每个边界都有不同的责任方和失败处理方式。
Distinguish “explain P/E ratio” from “what is the price now?” Current, latest, today, pre-market, after-hours, open, close, and market-status language should make the quote function available. Historical education questions may not need a live call.区分“解释市盈率”和“现在股价是多少”。当前、最新、今天、盘前、盘后、开盘、收盘和市场状态等表达应触发行情函数；历史知识类问题未必需要实时调用。
Use enums and narrow descriptions where the market universe is known. Treat Gemini’s function arguments as untrusted input: trim symbols, enforce list limits, resolve venue identifiers, and reject unexpected fields before any provider request.如果已知市场范围，应使用枚举和边界清晰的描述。Gemini 生成的函数参数仍是不可信输入：调用数据源前要清理代码、限制列表数量、解析市场标识并拒绝异常字段。
The backend selects credentials, endpoint, timeout, cache policy, retry count, and fallback. Those decisions should not be delegated to natural-language reasoning. Record which provider and entitlement produced the value.后端决定凭据、端点、超时、缓存、重试次数和降级策略，这些决策不应交给自然语言推理。还要记录具体数据源和授权层级。
Send Gemini only the fields needed to answer the question, plus traceability metadata. Keep raw provider payloads in application logs when permitted. Large trade histories, full order books, and redundant fields increase cost and make omissions harder to notice.只把回答所需字段和追踪元数据发给 Gemini。在许可范围内，把原始数据源载荷保存在应用日志中。庞大的成交历史、完整订单簿和冗余字段会增加成本，也更容易掩盖关键缺失。
Require the answer to show the resolved instrument, value, currency, timestamp with timezone, session, source, and delay label. If any required evidence is missing, Gemini should say what cannot be verified instead of repairing the gap with model knowledge.要求回答展示已解析标的、数值、币种、带时区时间戳、交易时段、来源和延迟标签。如果关键证据缺失，Gemini 应说明无法验证的部分，而不是用模型知识补齐空白。
Link the user request, Gemini function call, validated parameters, provider response ID, retrieval time, final answer, and error state. Redact secrets and sensitive portfolio details. This trace is what lets an operator explain a disputed quote later.关联用户请求、Gemini 函数调用、校验后参数、数据源响应 ID、获取时间、最终回答和错误状态；同时隐藏密钥和敏感持仓信息。这条轨迹使运维人员日后可以解释有争议的行情。
| Control控制项 | Implementation pattern实现方式 | Decision to document需要记录的决策 |
|---|---|---|
| Freshness新鲜度 | Set a maximum acceptable age per intent and market session.按意图和交易时段设置最大可接受数据年龄。 | When may cache be used, and how is cache age disclosed?何时允许使用缓存，怎样披露缓存年龄？ |
| Latency延迟 | Use bounded timeouts and one controlled retry for transient failures.设置明确超时，对暂时性故障只进行有限重试。 | Does Gemini wait, return partial evidence, or state failure?Gemini 应等待、返回部分证据，还是明确失败？ |
| Cost成本 | Cap symbols, deduplicate calls, batch safely, and monitor calls per conversation.限制股票数量、去重调用、安全批处理并监控每次对话的调用数。 | Which requests require confirmation before a broad scan?哪些大范围扫描必须先获得确认？ |
| Fallback降级 | Normalize alternate providers but preserve their identity and delay class.规范化备用数据源，同时保留其身份与延迟类别。 | Which differences make two feeds non-equivalent?哪些差异意味着两个 Feed 不能等同？ |
| Data rights数据权利 | Scope use by display, caching, derived output, logging, and redistribution terms.按展示、缓存、衍生输出、日志和再分发条款限制使用。 | May the Gemini answer be shown externally or stored?Gemini 的回答能否对外展示或保存？ |
Most failures happen outside model intelligence: ambiguous identity, incomplete metadata, incorrect entitlement assumptions, or an application that accepts a malformed tool result.多数问题并不是模型智力不足，而是标的身份歧义、元数据缺失、授权假设错误，或应用接受了异常的工具结果。
The function description may not mention current-price intent, or the application may allow a normal text response when a live call is mandatory. Add explicit trigger cases to tests and reject unsupported current-price answers at the application layer.函数描述可能没有覆盖当前价格意图，或者应用在必须实时查询时仍接受普通文本回答。应在测试中加入明确触发用例，并由应用层拒绝没有数据支持的当前价格回答。
Ticker strings are not globally unique. Resolve exchange, country, asset type, and share class. When confidence is low, return candidates to Gemini so it can ask the user a precise clarification question.股票代码并非全球唯一。应解析交易所、国家、资产类型和股类。置信度不足时，把候选项返回 Gemini，让它提出明确的澄清问题。
Providers may return a valid value even when the market is closed or the account lacks a live feed. Preserve the field’s meaning and session. Never infer freshness merely because the HTTP request completed now.休市或账户没有实时 Feed 时，数据源仍可能返回有效数值。必须保留字段含义与交易时段，不能因为 HTTP 请求刚刚完成就推断数据是实时的。
Convert timeout, rate limit, not-entitled, invalid-symbol, and empty-result states into explicit function errors. The final Gemini instruction should prohibit substituting memory when any current-data request fails.把超时、限流、未授权、无效代码和空结果转换为明确的函数错误。最终 Gemini 指令应禁止在当前数据请求失败时用记忆替代。
“What is AAPL trading at now?” “Latest TSLA price.” “How did NVDA move after hours?” Verify that Gemini requests a tool instead of answering from memory.“AAPL 现在多少钱？”“TSLA 最新价格。”“NVDA 盘后怎么走？”应验证 Gemini 发起工具调用，而不是凭记忆回答。
Test duplicate tickers, company names without an exchange, and prompts such as “that stock.” Gemini should ask for the missing identity rather than guess.测试重复代码、缺少交易所的公司名称以及“那只股票”等表达。Gemini 应询问缺失信息，不能猜测。
Simulate timeout, rate limit, missing entitlement, market halt, and a result without a timestamp. The answer must name the limitation and must not manufacture a quote.模拟超时、限流、未授权、停牌和缺少时间戳的结果。回答必须说明限制，不能编造价格。
Pass only when the final Gemini answer preserves the provider’s symbol, exchange, price, currency, timestamp, session, source, and delay status exactly—and labels hypothetical examples as examples.只有最终回答准确保留数据源返回的股票代码、交易所、价格、币种、时间戳、交易时段、来源与延迟状态，并明确标注假设示例时，才算通过。
A successful demo with one liquid U.S. ticker is not a release test. Use a compact, repeatable suite that covers identity, market time, provider behavior, and the model’s final wording.用一只高流动性美股完成演示并不等于可以上线。应使用精简、可重复的测试集，覆盖标的身份、市场时间、数据源行为与模型最终表达。
Use Gemini API function calling. Your backend controls credentials, execution, caching, and logging.使用 Gemini API 函数调用，由后端控制凭据、执行、缓存和日志。
Use a reviewed MCP server and expose only the quote tools required by the workflow.使用经过检查的 MCP Server，并且只开放工作流需要的行情工具。
Keep WebSocket ingestion outside Gemini. Aggregate the stream and call the model only for selected events or summaries.把 WebSocket 摄取留在 Gemini 之外。先聚合数据流，只在选定事件或摘要场景调用模型。
Do not treat Gemini itself as a licensed market-data feed. For a verifiable current quote, connect an external provider through function calling or MCP and disclose its source and timestamp.不要把 Gemini 本身当作具备授权的行情 Feed。若需要可验证的当前报价，应通过函数调用或 MCP 连接外部数据源，并披露来源和时间戳。
Use function calling when you own an application backend. Use MCP when Gemini CLI or another MCP-capable client needs to discover and call the market tool interactively.如果你控制应用后端，使用函数调用；如果 Gemini CLI 或其他 MCP 客户端需要交互式发现和调用行情工具，则使用 MCP。
A production system should ingest, filter, and aggregate the stream outside the model. Send Gemini a bounded snapshot or event payload rather than every tick.生产系统应在模型之外摄取、过滤和聚合数据流，再向 Gemini 发送有限的快照或事件载荷，而不是每个 Tick。
Keep this guide’s quote tool read-only. Trading requires a separate action tool, stronger authentication, deterministic risk checks, idempotency, audit logs, and explicit human confirmation.本指南中的报价工具应保持只读。交易需要独立动作工具、更强认证、确定性风险检查、幂等控制、审计日志和明确的人工确认。
Start with one symbol and a read-only tool. Confirm the selected capability, inspect its parameters, and reject any response that lacks source or freshness metadata.从一个股票代码和一个只读工具开始。确认所选能力、检查参数，并拒绝任何缺少来源或新鲜度元数据的结果。