DSers AliExpress 1688 Shopify Wix WooCommerce MCP TypeScript Smithery AI Agent Batch
2026.03

AI 一句话铺货 / DSers MCP Product

基于 DSers 平台的 AI 商品导入工具 — 用一句话完成选品-编辑-推送全流程
AI-powered product import tool built on DSers — one sentence replaces the entire import-edit-push workflow
Built on DSers · TypeScript · MCP · Smithery · 2026.03

一、DSers 市场地位 / DSers Market Position

DSers 是 AliExpress 官方 dropshipping 合作伙伴(Official Partner),2021 年 Oberlo 关停后成为行业标准工具。

DSers is the official AliExpress dropshipping partner. After Oberlo shut down in 2021, it became the industry standard.

500K+ 创业者在用 / Entrepreneurs
100M+ 管理商品数 / Products managed
$50M+ 年交易额 / Annual GMV

DSers 已经解决了 dropshipping 中最繁重的几个环节:

痛点 / Pain PointDSers 解决方案 / Solution
一单一单去速卖通下单 / Placing orders one by one批量下单,几百单几秒钟 / Bulk orders in seconds
手动同步物流单号 / Manual tracking sync自动同步 / Auto-sync with supplier
供应商断货 / Supplier out of stockSupplier Optimizer 一键替换 / One-click swap
多店铺来回切换 / Switching between stores统一后台 / Unified dashboard

二、未解决的痛点 / The Remaining Gap

DSers 在下单、物流、供应商管理方面做得很好,但选品-编辑-推送这一环节仍然是手动操作:

DSers handles orders, logistics, and suppliers well. But product curation — import, edit, push — is still fully manual:

找到商品 → 点导入 → 改标题(去中文、去水印文字)→ 改描述 → 调价格(乘倍率/按规则)→ 删变体 → 选图片 → 推送到 Shopify → 设置上架 → 等结果

Find product → Import → Edit title → Edit description → Adjust price → Remove variants → Pick images → Push to Shopify → Set visibility → Wait

当前流程 / Current

每个商品 10-15 次点击

每天上新 20 个 = 200-300 次点击

判断性操作无法批量化

CSV 批量导入解决"量",不解决"判断"

使用 MCP 后 / With MCP

每个商品 1 句话

每天上新 20 个 = 20 句话

AI 理解自然语言指令并自动执行

标题、价格、图片、上架全部可配置

三、解决方案 / Solution

将 DSers 的商品导入全链路封装为 7 个 MCP(Model Context Protocol)工具,AI agent 通过自然语言指令调用,替代手动点击:

DSers' entire product import workflow wrapped into 7 MCP tools. AI agents call them via natural language, replacing manual clicks:

MCP 工具功能 / Function替代操作 / Replaces
dsers.product.import解析商品链接(单条/批量),应用规则,生成预览找商品 + 导入 + 编辑
dsers.product.preview推送前预览变更检查编辑结果
dsers.store.push推送到店铺(单条/批量/多店铺)推送 + 发布设置
dsers.rules.validate校验定价/内容规则价格规则设置
dsers.store.discover查询店铺(含运费模板)和可用规则(新增能力)
dsers.product.visibility上架/下架控制可见性设置
dsers.job.status查询任务状态等待推送结果

架构特点:内容编辑(prepare 阶段)与发布行为(confirm 阶段)解耦,规则在 prepare 时冻结为快照,避免 AI 混淆内容参数和发布参数。

Architecture: Content editing (prepare) is decoupled from publishing (confirm). Rules are frozen as snapshots at prepare time, preventing AI from conflating content parameters with publish parameters.

四、一句话 = 一个场景 / One Sentence = One Scenario

每个业务场景对应一句标准话术,AI 自动翻译为正确的工具调用:

Each business scenario maps to one standard prompt. AI translates it into the right tool calls:

场景 / Scenario话术示例 / Prompt
快速铺货
Quick import
"导入这个商品到我的店铺,价格乘 3,先不上架。"
"Import this product to my store, 3x the price, don't publish yet."
https://aliexpress.com/item/100500xxx.html
精选上架
Curated listing
"导入这个速卖通商品,只保留前 5 张图,标题翻译成英文,描述只留 specifications,直接上架。"
"Import, keep first 5 images, translate title to English, specs only, publish directly."
高利润定价
High-margin
"导入这个商品,价格乘 4,用店铺定价规则,自动同步库存。"
"Import, 4x price, use store pricing rule, auto-sync inventory."
测品
Test product
"导入到后台,不发布、不上架,我先看看数据。"
"Import to backend only, don't publish, I want to check data first."
批量铺货
Bulk import
"把这 5 个链接全部导入,价格统一乘 2.5,全部先放后台。"
"Import all 5 links, uniform 2.5x pricing, all to backend."
多店铺同步
Multi-store
"导入这个商品,同时推送到我的 Shopify 和 Wix 店铺。"
"Import and push to both my Shopify and Wix stores."
混合来源批量
Mixed source
"导入这些链接(速卖通 + 1688 混合),把能导入的全部推送。"
"Import these links (AliExpress + 1688 mixed), push all valid ones."
指定运费模板
Shipping profile
"推送这个商品,使用 'General profile' 而不是默认的运费模板。"
"Push using 'General profile' instead of the default shipping profile."
推送前检查
Pre-push preview
"预览推送后的效果,价格和图片变了什么?"
"Preview post-push result — what changed?"
调整可见性
Visibility
"把这个商品从后台改成上架到在线商店。"
"Change from backend-only to published."
失败重试
Retry
"上次推送失败了,再推送一次到同一个店铺。"
"The last push failed, try pushing again to the same store."

五、技术架构 / Technical Architecture

项目 / Item说明 / Description
协议 / ProtocolMCP (Model Context Protocol) — AI agent 标准工具协议
语言 / LanguageTypeScript (Node.js 18+)
依赖 / Dependencies@modelcontextprotocol/sdk, zod, @smithery/sdk
商品来源 / SourceAliExpress, Alibaba, 1688
推送目标 / TargetShopify, Wix, WooCommerce(通过 DSers / via DSers)
发布平台 / RegistrySmithery — 在线连接或从 GitHub 本地安装 / Online connect or local install from GitHub
架构模式 / Pattern公开工作流契约 + 私有平台适配器 / Public workflow contracts + private platform adapters
状态管理 / StateServerless job recovery — 状态编码进 job_id,跨实例恢复 / State encoded in job_id, cross-instance recovery
托管 / HostingVercel (Streamable HTTP) + Smithery registry
安全设计 / Security不暴露平台 API 路径、认证方式、原始请求结构 / No exposed API paths, auth methods, or raw request structures

工程亮点 / Engineering Highlights

特性 / Feature说明 / Description
Dot-Notation Naming全部工具和 Prompt 采用 dsers.{domain}.{action} 分层命名(如 dsers.product.importdsers.store.push),AI agent 可自动发现层级关系
All tools and prompts use hierarchical dsers.{domain}.{action} naming (e.g. dsers.product.import, dsers.store.push) so AI agents can navigate the tool tree
MCP Annotations7 个工具全部标注 readOnlyHintdestructiveHintidempotentHintopenWorldHint,AI agent 可据此判断工具安全性
All 7 tools annotated with safety hints so AI agents can assess risk before calling
AI-Friendly Errors所有错误返回结构化 {Error, Cause, Action} 三段式,覆盖 15+ 错误场景
All errors return structured {Error, Cause, Action} covering 15+ error patterns
Prompt Templates3 个预设工作流模板:dsers.workflow.quick-import(快速导入)/ dsers.workflow.bulk-import(批量加价)/ dsers.workflow.multi-push(多店铺推送),MCP 客户端可直接展示
3 built-in prompts: dsers.workflow.quick-import / dsers.workflow.bulk-import / dsers.workflow.multi-push for instant MCP client display
大数字精度保护DSers 19 位 ID 全链路 string 化处理,避免 JavaScript Number 精度丢失
19-digit DSers IDs handled as strings throughout to prevent JavaScript precision loss
标题去重规则引擎自动检测已有前缀/后缀,防止重复应用
Rule engine auto-detects existing prefix/suffix to prevent duplication on re-import
全球 ID 解析AliExpress .us URL 自动提取 afTraceInfo 全球产品 ID
AliExpress .us URLs auto-resolve global product ID from afTraceInfo

六、开发背景 / Background

平台没有公开的 API 文档。所有接口都是我从网页里抓包分析出来的——打开浏览器开发者工具,一个请求一个请求地看,猜参数、试字段、反复验证。

The platform has no public API documentation. Every interface was reverse-engineered from the web page — browser DevTools open, inspecting requests one by one, guessing parameters, testing fields, verifying repeatedly.

拿到这些接口之后,我试过直接封装给 AI 用。结果 AI 推送商品到 Shopify 的时候连续失败了十几次。每次都报错,但每次错误原因都不一样。AI 每次都很认真地构造请求,每次都"看起来"是对的。但就是过不了。

After capturing these interfaces, I tried wrapping them directly for AI. The AI failed to push products to Shopify over a dozen times in a row — different error each time. AI carefully constructed requests every time, everything "looked" right, but it never went through.

最后发现:接口之间有隐含的前置依赖,抓包里看不出来,参数格式也和表面看到的不一致。AI 不可能猜到这些。

Turned out there were hidden dependencies between interfaces — invisible in network captures, parameter formats inconsistent with what they appeared to be. AI had no way to guess these.

AI 不笨,它只是被丢到了一个没铺好路的环境里。所以我做了这个 MCP:不是把抓来的接口直接扔给 AI,而是把坑都填好,把路铺平,让 AI 只需要走完工作流就行。

AI isn't stupid — it was just dropped into an environment with no paved road. So I built this MCP: instead of throwing captured interfaces at AI, I filled every pothole and paved the path, so AI just needs to follow the workflow.

七、项目状态 / Project Status

Released — Open Source

完整导入-编辑-推送链路已跑通。7 个工具 + 3 个 prompt 模板,全部带 MCP annotations,支持批量导入、多店铺推送(Shopify + Wix)、多来源混合、Shopify 运费模板自动发现、AI 友好错误提示。服务托管在 Vercel,已发布到 Smithery(Free 计划 25K RPCs/月),也可从 GitHub 克隆本地运行。同时提供 Python 版本(维护模式)。

Full import-edit-push pipeline verified. 7 tools + 3 prompt templates, all with MCP annotations. Supports batch import, multi-store push (Shopify + Wix), mixed sources, Shopify shipping profile auto-discovery, and AI-friendly structured error messages. Hosted on Vercel, published on Smithery (Free plan 25K RPCs/month), or clone from GitHub to run locally. A Python version is also available (maintenance mode).