跳到正文
DEV Community· Khairul Fauzy Khairudin·· 11 天前AI 評分67

作者建立 MCP 伺服器,讓 Claude 即時讀取 App Store 和 Google Play 評價

I gave Claude live App Store and Google Play reviews with one MCP server

AI 導讀

作者建立並透過 Apify 提供 MCP 伺服器,讓 Claude 即時搜尋 App Store 和 Google Play 的應用程式及評價,並整理評分、好評與投訴等資訊。

正文

Khairul Fauzy Khairudin

Ask Claude "what do users hate about Todoist on Android right now?" and it will guess from old training data. I wanted real answers from this week's reviews, so I built an MCP server that reads both app stores live and hands the model a compact summary instead of a pile of raw text.

What it does

Five tools, one server:

Tool What the agent gets
search_apps Apps on App Store, Google Play or both, with ratings and ids
get_app_details Full listing: rating histogram, installs, version, release notes
get_app_reviews Raw reviews, filterable by star rating (e.g. only 1–2 stars)
analyze_app_reviews Insights report: complaint topics, praise topics, rating by version and month, developer reply rate, short quotes
compare_apps The same report for 2–5 competitors side by side

You can pass an app as a store URL, an Apple id, an Android package name, or just a name.

Why a summary tool matters

Raw reviews are expensive for an agent. 300 reviews can easily fill tens of thousands of tokens, and the model still has to count topics by hand. analyze_app_reviews does the counting first and returns about 4–5 KB of JSON.

Here is a real call on Todoist's 100 latest US Google Play reviews (shortened):

{
  "averageRating": 3.93,
  "ratingDistribution": { "1": 14, "2": 7, "3": 9, "4": 12, "5": 58 },
  "negativeShare": 0.21,
  "developerReplyRate": 0.85,
  "topComplaints": [
    { "term": "task", "mentions": 5, "shareOfReviews": 0.238 },
    { "term": "add", "mentions": 4, "shareOfReviews": 0.19 }
  ]
}

The store shows Todoist at 4.6 stars overall. The latest 100 reviews average 3.93, and 21% are 1–2 stars. That gap is exactly what a product manager wants to see, and it only shows up in fresh data.

Connect it

It runs on Apify as a Standby actor, so there is nothing to host. In Claude Desktop, Cursor or VS Code:

{
  "mcpServers": {
    "app-reviews": {
      "url": "https://rasareka--app-reviews-mcp.apify.actor/mcp",
      "headers": { "Authorization": "Bearer YOUR_APIFY_TOKEN" }
    }
  }
}

Claude Code:

claude mcp add --transport http app-reviews https://rasareka--app-reviews-mcp.apify.actor/mcp --header "Authorization: Bearer YOUR_APIFY_TOKEN"

Prompts that work well

  • "Compare the latest reviews of Todoist, TickTick and Things 3. Where is each one weakest?"
  • "Show me 1-star Notion reviews from the UK App Store since the last update."
  • "Find the top 10 habit tracker apps in the US and tell me which has the most complaints about sync."

Pricing and limits

Pay per use: $0.005 per search or details call, $0.005 per 50 reviews, $0.03 per insights report. Failed calls are free. A "compare 3 competitors" session costs about $0.10.

Honest limits: Apple's public feed stops at 500 reviews per app per country, and topic extraction is keyword-based, so the agent should read the quotes for nuance.

Actor page: https://apify.com/rasareka/app-reviews-mcp

I'd love feedback on which tool you would add next.

來源:DEV Community · dev.to