# Asklear Agent Docs / Asklear Agent 使用手册

Asklear connects verified clients such as Codex, Claude Code, WorkBuddy, and Doubao Work, plus enterprise-built Agents, through MCP. The ChatGPT desktop app has a trial connection recipe pending host publication and live routing acceptance. Asklear helps AI agents acquire in-depth web content and commercial data for market analysis, category research, competitor comparisons, user research, and trend analysis. / Asklear 通过 MCP 接入 Codex、Claude Code、WorkBuddy、豆包工作等已验证客户端和企业自建 Agent。ChatGPT 桌面客户端目前提供试用接入指引，等待宿主发布与真实路由验收。Asklear 为 Agent 提供商业研究所需的深度网页内容和商业数据，适用于市场分析、品类研究、竞品对比、用户研究与热点分析。

## What Asklear can do / Asklear 能做什么

- **Historical sales data / 历史销售数据**: Access product information, sales volume, sales revenue, prices, and historical trends from major e-commerce platforms in China, including Tmall, JD.com, Douyin E-commerce, and Pinduoduo, for market analysis, category research, and competitor comparisons. / 获取中国主流电商平台的商品信息、销量、销售额、价格与历史趋势，包括天猫、京东、抖音电商、拼多多等平台，用于市场分析、品类研究和竞品对比。
- **Current platform content / 平台最新内容**: Search major social media and content platforms in China, including Xiaohongshu, Douyin, Weibo, Bilibili, and WeChat Official Accounts, and retrieve the latest content, details, and engagement data for user research, trend analysis, and topic research. / 搜索中国主流社交媒体与内容平台，包括小红书、抖音、微博、哔哩哔哩、微信公众号等，能够获取最新内容、详情及互动数据，用于用户研究、热点分析和话题研究。
- **Cloud web collection / 云端网页采集**: Read public webpage URLs in batches or search them with Google Advanced and Bing Basic, then return clean, readable content. It can also serve as a fallback when an Agent cannot access a webpage locally. / 批量读取公开网页链接，或使用 Google Advanced、Bing Basic 搜索公开网页，并返回干净、便于阅读的内容；也可在 Agent 本地无法访问网页时作为备用获取方式。
- **Local browser collection / 本地浏览器采集**: Let an Agent read and interact with pages in the user's signed-in browser, for tasks that require a login session or local browser environment. / 让 Agent 在用户已登录的浏览器中读取和操作页面，适合需要登录状态或本地浏览器环境的任务。

Connect Asklear before your first task. Doubao Work can use the URL-only HTTP + OAuth path described by Agent Setup. The ChatGPT desktop app is currently an unverified trial: use the Asklear plugin only when it is available in your workspace, or follow the Developer mode MCP connection steps for pre-publication testing. Once connected, ask the original business question directly. / 第一次使用时，请先连接 Asklear。豆包工作优先使用 Agent Setup 提供的仅 URL HTTP + OAuth 路径。ChatGPT 桌面客户端目前是未验证试用入口：工作区已有 Asklear Plugin 时才安装或启用，发布前测试则按 Agent Setup 使用 Developer mode 添加 MCP 连接。连接完成后直接提出原始业务问题。

> Canonical Agent entry for capability routing, query execution, research methods, and error recovery.
> 能力路由、查询执行、研究方法和错误恢复的唯一 Agent 入口。
> Read only the relevant page, then return to the user's original business task.
> 只读取当前任务需要的页面，然后返回用户最初的商业任务。

## Contents / 目录

- [CLI integration / CLI 接入指南](https://docs.asklear.cn/cli): `agent.cli` · Connect scripts and agents to Asklear through the command line. Discover, quote, start and retrieve tasks without configuring MCP. / 用命令行连接 Asklear，发现任务、获取报价、执行采集并读取结果。适用于脚本和 Agent；不需要先配置 MCP。
- [Capabilities & Data / 能力与数据](https://docs.asklear.cn/capabilities): `agent.capabilities` · Choose the evidence Asklear can provide for the business question. / 根据商业问题选择 Asklear 可以提供的证据。
- [Query Guide / 查询指南](https://docs.asklear.cn/querying): `agent.querying` · Use tools/list as the authoritative tool surface and take the shortest safe route: a known exact reference or resolved value goes to query_metrics, a product name or title goes through search_products, and an unresolved brand, shop, or category goes through search_values first. connection_status is free and not a heartbeat; it checks Asklear connection only. Keep remote data MCP, async collection, and local browser MCP separate. Explain Asklear billing metadata and never expose provider fields or fees. / 以 tools/list 作为权威工具面并走最短安全路径：已知精确引用或已解析值直接 query_metrics，商品名称或标题先用 search_products，未解析的品牌、店铺或品类先用 search_values。connection_status 免费且不是心跳，只检查 Asklear 连接。远程 data MCP、异步采集和本地浏览器 MCP 分开。解释 Asklear 计费 metadata，绝不暴露 provider fields 或费用。
- [Agent Task Migration / Agent 任务迁移](https://docs.asklear.cn/workflows): `agent.workflows` · Use the high-level Agent task workflow while keeping fine-grained Data APIs available to developers. / 面向 Agent 使用高层任务流程，同时为开发者保留完整细粒度 Data API。
- [Browser Extraction / 浏览器提取](https://docs.asklear.cn/browser): `agent.browser` · Use the local Asklear browser MCP when an Agent must work in the user's signed-in Chrome. It is independent from public web collection and sales-data queries; page content and login state stay local. / 当 Agent 需要在用户已登录的 Chrome 中工作时，使用 Asklear 本地浏览器 MCP。它独立于公开网页采集和销售数据查询；页面内容与登录态留在本地。
- [Cookbook / 研究方法](https://docs.asklear.cn/cookbook): `agent.cookbook` · Reusable ways to turn a business question into an Asklear research task. / 将商业问题转化为 Asklear 研究任务的可复用方法。
- [Errors & Recovery / 错误与恢复](https://docs.asklear.cn/errors): `agent.errors` · Find the recovery action for an Asklear error. / 根据 Asklear 错误找到明确的恢复动作。

## Datasets / 数据集

- [Douyin Monthly Product Sales / 抖音商品月度销量明细表](https://docs.asklear.cn/capabilities/douyin): Douyin / 抖音 → Douyin Monthly Product Sales / 抖音商品月度销量明细表 → `dataset=douyin`
- [JD Monthly Product Sales / 京东商品月度销量明细表](https://docs.asklear.cn/capabilities/jd): JD / 京东 → JD Monthly Product Sales / 京东商品月度销量明细表 → `dataset=jd`
- [PDD Monthly Product Sales / 拼多多商品月度销量明细表](https://docs.asklear.cn/capabilities/pdd): PDD / 拼多多 → PDD Monthly Product Sales / 拼多多商品月度销量明细表 → `dataset=pdd`
- [Tmall Monthly Product Sales / 天猫商品月度销量明细表](https://docs.asklear.cn/capabilities/tmall): Tmall / 天猫 → Tmall Monthly Product Sales / 天猫商品月度销量明细表 → `dataset=tmall`

## Capability routing / 能力路由

- The question concerns a market, category, competitor, product, sales, or revenue. / 问题涉及市场、品类、竞品、商品、销量或销售额。 → Use the relevant platform dataset; read its contract only when fields or definitions are uncertain. / 使用相关平台数据集；仅在字段或口径不确定时读取其契约。
- The answer requires current social or content-platform search results, details, or engagement data. / 答案需要当前社媒或内容平台的搜索结果、详情或互动数据。 → Use the matching platform task contract and collection workflow. / 使用匹配的平台任务契约和采集流程。
- The task needs Google Advanced or Bing Basic public-web search, multiple public URLs, or a known page the Agent cannot fetch locally. / 任务需要 Google Advanced 或 Bing Basic 公开网页搜索、批量读取多个公开 URL，或 Agent 本地无法读取某个网页。 → Use the cloud web-collection task and its collection workflow. / 使用云端网页采集任务及其采集流程。
- The page requires login, clicking, filling, scrolling, or pagination in the user's browser. / 页面需要在用户浏览器中登录、点击、填写、滚动或翻页。 → Use the local browser MCP and read agent.browser when its workflow is uncertain. / 使用本地浏览器 MCP；流程不确定时读取 agent.browser。

## Runtime authority / 运行时准据

- Authenticated tools and runtime responses are authoritative for current access, coverage, fields, and billing. / 当前权限、覆盖、字段和计费以认证后的工具面与运行时响应为准。
- Provide and reuse the exact `task_query` when possible; omitted values use a fixed non-user-specific attribution. / 尽量提供并复用同一任务的 `task_query`；未提供时使用固定的非用户归因。