[中文]

用途:

• 微信「搜一搜」综合搜索，垂类由 business_type 字符串键切换，传关键词即可。
• 实测可搜：综合 / 公众号（含服务号、视频号）/ 文章 / 视频号视频 / 表情。
• ⚠️ 其余垂类（直播 / 朋友圈 / 新闻 / 读书 / 听书 / 图片 / 百科 / 微信指数 / 划线 / 评论 / 贴图 / 小游戏 / AI 搜索）由微信服务端按账号与地区下发，当前实测恒返回空结果（no_more 文案非空即为正常空结果，raw=true / raw=false 都在顶层；raw=false 时另有 count=0 + items=[]。仍按一次正常调用计费）。
• 价格：0.01$/次
• ⏱️ 由于微信服务器原因，本接口响应较慢，请将客户端请求超时（timeout）设置为 30 秒；timeout 设置过小会造成已扣费但收不到响应的情况。
• ⚠️ 大整数 ID 精度：响应中的 docID / feedNonceId 等为 64 位大整数，超出 JavaScript 安全整数范围（2^53-1）。请始终以字符串方式接收 / 传递这类 ID（解析 JSON 用 json-bigint 或按文本取值），勿让其经过 JS Number。Swagger UI 文档页对超大整数会末位舍入显示，属正常现象，不影响接口实际返回的数据。

参数:

• keyword: 搜索关键词（去空白后 1-100 字）。示例 人民日报
• business_type: 可选，默认 all。垂类字符串键（共 18 个）。实测有结果：综合 all / 公众号·服务号·视频号 account / 文章 article / 视频 video / 表情 sticker。其余键 underline / encyclopedia / live_stream / comment / listen / news / photos / book / moments / image / mini_game / weixin_index / ai_search 由服务端按账号与地区下发，当前实测恒空。传列表外的值将直接报错（422）。
• sort: 可选，默认 default。排序（结果页「排序」下拉，综合与各垂类通用）—— 不限(相关性) default/0 / 最新(发布时间降序) latest/1 / 最热(点赞降序) hot/2，字符串键或整数均可，非法值报错（400）。
• publish_time: 可选，默认 all。发布时间（结果页「时间」下拉，通用）—— 不限 all/0 / 最近一天 day/1 / 最近七天 week/2 / 最近半年 half_year/3，字符串键或整数均可，非法值报错（400）。
• offset: 可选，默认 0（>=0）。首页传 0；翻页请用 cursor，只传 offset 无效（每页都回到第一页）。
• cursor: 可选。翻页游标。首页留空；翻页时把上一页响应里返回的 cursor 原样传回（须同时重复上一页的 sort / publish_time），配合 continue_flag 判断是否还有下一页。
• raw: 可选，默认 True。True=原始搜索响应；False=精简解析。

💡 视频垂类专属的「时长」筛选见 /fetch_search_videos（business_type=video + duration）。

返回:

• 搜索结果列表（结构随垂类略有差异）

典型链路:

• 💡 拿视频号详情（媒体下载地址 / decode_key）：搜索（video / all）的视频结果项不直接含媒体，只给 exportId（+ jumpInfo.extInfo.feedNonceId）。要下载 / 解密：取 exportId → 走视频号 V2 接口 /api/v1/wechat_channels/v2/fetch_video_detail（传 export_id）拿 media（url / url_token / decode_key）和明文 username。完整链路与解密说明见该接口文档。
• 公众号结果项的 jumpInfo.userName（gh_…）可喂给公众号 V2 接口 /api/v1/wechat_mp/v2/fetch_account_profile / fetch_account_articles。

响应结构与 JSON Path:

raw=false（精简解析，snake_case）:

• 关键词回填: $.data.keyword
• 实际生效的垂类代码（如公众号=33554499）: $.data.business_type
• 结果总数: $.data.total —— 响应里所有 box 与 subBox 的 totalCount 全扫取最大值（垂类的 totalCount 挂在 subBoxes 上；综合搜索没有单一总数，给的是最大的那个 box）。零结果时为 null（没有任何正数 totalCount），恒空垂类恒为 null
• 是否还有下一页: $.data.continue_flag
• 翻页游标（原样回传接口的 cursor 参数取下一页）: $.data.cursor
• 服务端 offset（仅供参考，单独传回无法翻页）: $.data.offset
• 本页结果数: $.data.count
• 拍平的结果项列表（已把各 box / subBox 的 items 合并）: $.data.items[]
• 服务端下发的垂类 tab 清单（仅综合 business_type=all 时有值，其余为 null）: $.data.categories[] —— 每项 .type / .word / .extra_kvs，这是当前账号真正可用垂类的权威来源。注意 .type 是垂类的整数码（如 33554499），而本接口的 business_type 只收上面那 18 个字符串键，整数会被拒（422）—— 需自行按键表映射（如 33554499 → account）
• 零结果时服务端给的文案: $.data.no_more（有值 = 微信返回的正常空结果，不是调用失败）
• 单项（N 为下标，字段随垂类不同；以下为 account 公众号示例）:
  • 标题（含 <em> 高亮）: $.data.items[N].title
  • 描述: $.data.items[N].desc
  • 文档 id: $.data.items[N].docID
  • 类型名（如「公众号」）: $.data.items[N].accTypeName
  • 跳转信息: $.data.items[N].jumpInfo（.userName = gh_…、.nickName、.signature）
  • 视频垂类项另含 exportId（+ jumpInfo.extInfo.feedNonceId）→ 见上「典型链路」

raw=true（原始，本接口默认）:

• data 顶层: keyword / business_type / results（不做 items 拍平）/ total / continue_flag / offset / cursor / categories / no_more（total / categories / no_more 两种模式都在顶层，字段集已对齐）
• 完整结果集: $.data.results
• 结果盒子数组: $.data.results.data[]，每个 box 的结果在 $.data.results.data[M].items[] 与 $.data.results.data[M].subBoxes[K].items[] —— 即 raw=false 的 items[] 是把这些拍平后的结果。
• 是否还有下一页: $.data.continue_flag（raw=true / raw=false 均在顶层）
• 翻页游标: $.data.cursor —— 原样回传接口的 cursor 参数取下一页；$.data.offset 仅为服务端 offset，单独传回无法翻页
• 总数直接读顶层 $.data.total（raw 两种模式都有）—— 它是 $.data.results.data[M].totalCount 与 $.data.results.data[M].subBoxes[K].totalCount 全扫取的最大值；垂类的数字挂在 subBoxes 上，只读顶层 box 的 totalCount 会拿到 null
• 做结果遍历建议直接用 raw=false（items 已拍平），路径更干净。

[English]

Purpose:

• WeChat "Search" (搜一搜) universal search; the vertical is switched by the business_type string key — just pass a keyword.
• Verified to return data: all / official accounts (incl. service accounts and Channels) / articles / Channels videos / stickers.
• ⚠️ The other verticals (live streams, Moments, news, books, listen, images, encyclopedia, WeChat Index, underline, comments, photos, mini games, AI search) are gated server-side per account & region and currently always come back empty. A non-empty no_more is the universal signal (top level in both raw=true and raw=false); raw=false additionally gives count=0 + items=[]. Still billed as a normal call.
• Price: $0.01 per request
• ⏱️ Due to WeChat server latency, this endpoint responds slowly; please set your client request timeout to 30 seconds — a timeout that is too small may result in being billed without receiving the response.
• ⚠️ Large-integer ID precision: IDs such as docID / feedNonceId in the response are 64-bit big integers beyond JavaScript's safe-integer range (2^53-1). Always receive / pass such IDs as strings (parse JSON with json-bigint or read them as text), never through JS Number. Swagger UI rounds the trailing digits of huge integers in its docs view — this is expected and does not affect the actual data returned by the API.

Parameters:

• keyword: Search keyword (1-100 chars after trimming). E.g. 人民日报
• business_type: Optional, default all. Vertical string key (18 total). Verified to return data: all / account / article / video / sticker. The rest (underline / encyclopedia / live_stream / comment / listen / news / photos / book / moments / image / mini_game / weixin_index / ai_search) are gated server-side per account & region and currently always come back empty. Any value outside this list is rejected (422).
• sort: Optional, default default. Sort (result page "sort" dropdown, common to all verticals) — default/0 (relevance) / latest/1 (newest) / hot/2 (most liked); string key or integer, invalid values rejected (400).
• publish_time: Optional, default all. Publish time (result page "time" dropdown, common) — all/0 / day/1 / week/2 / half_year/3; string key or integer, invalid values rejected (400).
• offset: Optional, default 0 (>=0). Pass 0 for the first page; use cursor to paginate — offset alone does not work (it returns the first page every time).
• cursor: Optional. Pagination cursor. Leave empty for the first page; for the next page pass back the cursor returned in the previous response (repeat the same sort / publish_time), and use continue_flag to check whether there are more pages.
• raw: Optional, default True. True=raw search response; False=simplified parsing.

💡 The "duration" filter specific to the video vertical is available at /fetch_search_videos (business_type=video + duration).

Return:

• Search result list (structure varies slightly by vertical)

Typical chains:

• 💡 To get Channels video detail (media download address / decode_key): video result items from search (video / all) do not contain media directly — only exportId (+ jumpInfo.extInfo.feedNonceId). To download / decrypt: take exportId → call the Channels V2 endpoint /api/v1/wechat_channels/v2/fetch_video_detail (pass export_id) to get media (url / url_token / decode_key) and the plain username. See that endpoint's docs for the full chain and decryption notes.
• The jumpInfo.userName (gh_…) of official account result items can be fed into the MP V2 endpoints /api/v1/wechat_mp/v2/fetch_account_profile / fetch_account_articles.

Response structure & JSON Path:

raw=false (simplified, snake_case):

• Keyword echo: $.data.keyword
• Effective vertical code (e.g. account=33554499): $.data.business_type
• Total results: $.data.total — the maximum totalCount scanned across every box and subBox (a vertical's totalCount hangs off subBoxes; the All tab has no single total, so this is the largest contributing box). null on a zero-result page (no positive totalCount anywhere), so it is always null for the always-empty verticals
• Has next page: $.data.continue_flag
• Pagination cursor (pass back as the cursor parameter for the next page): $.data.cursor
• Server-side offset (informational only; passing it back alone does not paginate): $.data.offset
• Results on this page: $.data.count
• Flattened result item list (items of all boxes / subBoxes merged): $.data.items[]
• Server-issued vertical tab list (only present for business_type=all, null otherwise): $.data.categories[] — each entry has .type / .word / .extra_kvs. This is the authoritative list of verticals actually available to the serving account. Note .type is the vertical's integer code (e.g. 33554499) while this endpoint's business_type accepts only the 18 string keys above — an integer is rejected (422), so map it yourself (e.g. 33554499 → account).
• Server wording for a zero-result page: $.data.no_more (present = a legitimate empty result from WeChat, not a failed call)
• Single item (N is the index; fields vary by vertical; below is an account example):
  • Title (with <em> highlight): $.data.items[N].title
  • Description: $.data.items[N].desc
  • Document id: $.data.items[N].docID
  • Type name (e.g. "公众号"): $.data.items[N].accTypeName
  • Jump info: $.data.items[N].jumpInfo (.userName = gh_…, .nickName, .signature)
  • Video vertical items also carry exportId (+ jumpInfo.extInfo.feedNonceId) → see "Typical chains" above

raw=true (raw, default for this endpoint):

• data top level: keyword / business_type / results (items are not flattened) / total / continue_flag / offset / cursor / categories / no_more (total / categories / no_more are top level in both modes — the key sets are aligned)
• Full result set: $.data.results
• Result box array: $.data.results.data[]; each box's results are in $.data.results.data[M].items[] and $.data.results.data[M].subBoxes[K].items[] — the items[] of raw=false is the flattened merge of these.
• Has next page: $.data.continue_flag (top level for both raw=true / raw=false)
• Pagination cursor: $.data.cursor — pass it back as the cursor parameter for the next page; $.data.offset is only the server-side offset and passing it back alone does not paginate
• Read the total from the top level $.data.total (present in both raw modes) — it is the maximum of $.data.results.data[M].totalCount and $.data.results.data[M].subBoxes[K].totalCount; a vertical's number hangs off subBoxes, so reading only the top-level box's totalCount yields null
• For result iteration, prefer raw=false (items already flattened) for cleaner paths.

{
  "keyword": "人民日报",
  "business_type": "account"
}