Shopee Price Data in Claude & Cursor (MCP Setup)

If you have ever asked ChatGPT or Claude “what is my competitor charging on Shopee right now”, you already know the answer: it does not know. Language models have no live access to marketplace data, and Shopee does not offer a public price API to third parties. So the model guesses, hedges, or invents a number.

MCP closes that gap. It is the mechanism that lets a model call a real data source instead of guessing — and once it is wired up, asking about competitor prices becomes as ordinary as asking about the weather.

This guide covers the whole path: what MCP actually is, the three-step setup, and the questions that make it worth the ten minutes.

Why your AI cannot see Shopee prices

Three separate obstacles, and each one matters:

1. No public data. Shopee has no open product/price API for sellers or third parties. Anything you see quoted in a chat is either stale or fabricated.
2. Rendering. Product pages are built client-side, so a naive HTTP request returns a shell without prices. Scraping reliably needs a browser context and a lot of maintenance.
3. Rate limits and risk. Hammering product pages from your own IP is exactly the pattern anti-bot systems are built to catch — and on a seller account, that risk is not worth taking.

MCP sidesteps all three by separating collecting data from asking questions about data. Collection runs on PriceMinder’s scheduled pipeline; questions go through a small, authenticated tool interface.

What MCP is, in one paragraph

Model Context Protocol is a standard way for an AI client to discover and call external tools. Think of it as USB-C for AI: the client does not need to know how a service works internally, only that the service exposes tools with names, descriptions and typed parameters. When you ask “which of my competitors dropped prices this week”, the model reads the available tool list, picks get_alerts or get_price_summary, calls it with the right arguments, and turns the JSON response into an answer.

The practical consequence: your AI gets a real data source, and you get answers with numbers in them.

Step 1 — Get your MCP key

Log in to the PriceMinder dashboard, open Settings, and generate an MCP key. It is shown once, next to a full connection URL:

https://priceminder.online/mcp_server?key=YOUR_MCP_KEY

Treat that URL as a password — it authenticates as you. Do not paste it into a repository, a screenshot or a shared chat. If it leaks, regenerate the key and the old one dies immediately.

Step 2 — Add the server to your client

The transport is Streamable HTTP, so there is no local process to install and no npx command to run. Your client connects straight to the hosted endpoint.

Claude Desktop — Settings → Developer → Edit Config:

{
  "mcpServers": {
    "shopee-sentinel": {
      "type": "streamable-http",
      "url": "https://priceminder.online/mcp_server?key=YOUR_MCP_KEY"
    }
  }
}

Cursor — Settings → MCP → Add new MCP server, type HTTP, paste the URL. Or commit .cursor/mcp.json at the project root so the whole team shares it (mind the key — use an environment variable if the repo is shared):

{
  "mcpServers": {
    "shopee-sentinel": {
      "type": "streamable-http",
      "url": "https://priceminder.online/mcp_server?key=YOUR_MCP_KEY"
    }
  }
}

Cherry Studio — Settings → MCP Servers → Add Server → type Streamable HTTP, paste the URL.

Cline — open the MCP marketplace tab, choose “Remote / HTTP server”, paste the URL.

Windsurf — Cascade → Settings → MCP → Add MCP Server, paste the URL.

Restart the client if the tool list does not refresh. A correctly configured server shows ten tools; if you see zero, the URL is almost always truncated at the ?key= part.

Step 3 — Verify in one sentence

Ask something only your own data can answer:

How many products am I currently monitoring?

If it returns a count matching your dashboard, the connection works. If it returns a generic “I don’t have access to that”, the tool call failed — re-check the key.

The ten tools, and what they are actually for

You never call these by name. You ask a question; the model picks. But knowing the surface tells you what is worth asking.

Tool	What it answers
get_monitor_list	What am I tracking, and is it active?
get_monitor_overview	Snapshot of one product: current price, rank, status
search_my_products	Find a monitor by keyword when the list is long
add_monitor	Add a product to the watchlist by URL or ID
update_monitor_status	Pause or resume monitoring for a product
get_price_history	The actual price series over a date range
get_price_summary	Aggregated movement: min, max, change rate, direction
get_alerts	Which prices crossed my threshold, and when
mark_alert_read	Clear an alert from the inbox
get_crawl_health	Is collection running normally for my sites?

Eight questions worth trying

These are the ones that change how a seller works, rather than just demonstrating that the plumbing is connected.

1. “Which competitors dropped price more than 5% in the last 7 days?” — the morning triage that normally takes twenty tab-switches.
2. “For the Bluetooth earbuds I track, give me the price range of the last 30 days and today’s position in it.” — instant pricing context before you set a promotion.
3. “Compare my price with each competitor on the same model and tell me who is cheapest.” — the price-gap view, in one line.
4. “Is collection healthy for the Thailand site? Any missed crawls this week?” — data trust before data decisions.
5. “Show me every alert from last week that I have not read.” — an alert inbox that respects your attention.
6. “Add this product to my watchlist: [URL].” — onboarding a competitor without opening the panel.
7. “Which of my monitored products have not changed price in 14 days?” — dead listings and stale competitors surface fast.
8. “Summarise the price war on this category in three bullets for my team.” — the model does the writing, the data comes from the tools.

Pair any of these with a follow-up — “and what should I do about it?” — and the answer improves when the Smart Follow decision tools are in scope, because the model can reason over the price gap, rating and sales momentum instead of price alone.

What this does not do

Worth stating plainly, because the alternative is disappointment:

• It does not scrape on demand. Data freshness is the crawl schedule, not the moment you ask. If you need minute-level freshness for a flash sale, raise the check frequency on Pro rather than expecting the AI to fetch live.
• It does not edit your listings. Reading, monitoring and pausing is the reachable surface; price changes stay manual.
• It does not replace judgement. An AI that can see five competitors’ prices still cannot see your inventory position or your margin floor. It shortens the loop from “something moved” to “here is the number” — the decision is still yours.
• It is not free forever at scale. Free covers the workflow with a quota; sustained daily use belongs on Pro.

Start with one question

The cheapest way to evaluate this is to not evaluate it — install the extension, add two competitors you already watch by hand, generate a key, and ask question one above tomorrow morning. If the answer is faster than your current routine, the rest follows.

• Install the browser extension — 4 steps, Chrome / Edge / Opera / 360
• Full MCP setup reference — screenshots and video, per client
• What Smart Follow does with the data — from price movement to a reprice call