TL;DR: EasyCatch runs a local-first Model Context Protocol (MCP) server on your own computer. It lets Claude Code, Cursor and other MCP clients read the 1688 product page open in your browser, including tiered prices, SKUs and spec parameters, through the EasyCatch Chrome extension. Nothing goes through a cloud scraper, and your 1688 login stays in your browser.
How does the EasyCatch 1688 MCP server work?
The server speaks MCP over stdio to your AI client and opens a WebSocket relay on 127.0.0.1:19876 for the EasyCatch Chrome extension. When the agent calls a tool, the request goes from the client to the local server, through the relay to the extension, and into the 1688 parser that runs inside the page. The parsed result travels back the same way. No hop leaves your machine except the page load you already made in Chrome.
The server exposes four tools:
| Tool | What it does | Needs the extension online |
|---|---|---|
easycatch_extract_current_tab | Extracts the active tab as Markdown; on 1688 it adds tiered prices, SKUs, spec parameters and main, SKU and detail images | Yes |
easycatch_fetch_url | Fetches a public URL from the server and converts it to Markdown | No |
easycatch_search_history | Searches your local capture history by keyword, domain or date | Yes |
easycatch_dispatch | Sends Markdown to a configured target such as Feishu, Obsidian, DingTalk or a webhook | Yes |
For pages that need a login, easycatch_extract_current_tab is the right tool, because it works on the session you already have open.
Local MCP Server vs cloud scraper API
| Dimension | EasyCatch Local MCP Server | Cloud Scraper API |
|---|---|---|
| Cost | Runs on your own machine, so there is no per-request bill; the MCP gateway is part of EasyCatch Pro ($9.90/month) | Per-request or monthly subscription pricing, plus the AI tokens spent cleaning raw HTML |
| Latency | Parses the page already loaded in your tab; the published EasyCatch CDN sniffing benchmark is 0.02 s | Every call adds a network round trip and a remote browser render |
| Store safety | Uses your own browser session; no password or cookie is handed to a third party | You share cookies or credentials with a proxy pool, which 1688 risk control can flag |
| Spec alignment | Same 1688 parser as the extension, so prices, SKUs and spec parameters come back structured | Returns raw HTML or generic JSON that you post-process yourself |
How do you connect an AI client?
- Install the EasyCatch Chrome extension and open its side panel, which starts the WebSocket bridge.
- Start the EasyCatch MCP server and register it in your MCP client config (Claude Code, Claude Desktop, Cursor).
- Open a 1688 product page in Chrome and ask the agent to extract the current tab.
The config panels and webhook payload formats are on the EasyCatch Developer Hub. To turn the captured product into a Shopify listing, follow the 1688 to Shopify import guide.
Frequently Asked Questions
What is a 1688 MCP server?
It is a Model Context Protocol server that lets an AI client such as Claude Code or Cursor call tools that read 1688 product pages. The EasyCatch MCP server runs on your own computer and talks to the EasyCatch Chrome extension, which parses the 1688 page you have open: title, tiered prices, SKUs, spec parameters and images.
Do I have to give my 1688 account password to the MCP server?
No. The extract tool reads the tab that is already open and logged in inside your own Chrome profile. The server only receives the parsed result over a local WebSocket bound to 127.0.0.1, so your 1688 login, cookies and password never leave the browser.
Can the MCP server export a Matrixify CSV for Shopify?
Not directly. The MCP tools return structured page content to the AI client and can dispatch it to Markdown, Feishu, Obsidian or a webhook. The Matrixify-compatible CSV and image ZIP are exported from the EasyCatch side panel, using the same 1688 parser.
Written by Zi Dong Liang
Founder & Lead Engineer, EasyCatch
Zi Dong Liang is the founder and lead engineer behind EasyCatch. He built the EasyCatch MCP gateway so AI coding agents can read 1688 product data from the browser session a sourcing team already uses, without a cloud scraper in the middle.