Review Retriever · setup

Connect app reviews to your MCP client.

Set this up once and the export step disappears. You paste a store URL into the chat and the reviews arrive — up to 500 of them, with stable evidence IDs so every claim in the analysis can be traced back to the sentence that produced it.

Review Retriever retrieves the evidence. Review Intelligence analyzes it. The connector does not log in to a store account, change store data, or turn review patterns into proof of revenue, retention, or causality — and the boundary below is specific about all three.

Serverhttps://reviews.doubledash.me/mcp

TransportRemote HTTP

Toolretrieve_app_reviews

AccessPublic reviews · read-only

Before you connect

You need an MCP client that supports remote HTTP servers, and a public App Store or Google Play URL. That is the whole list. No API key, no App Store Connect login, no Play Console login, no OAuth client — which also means there is nothing here for you to accidentally leak.

The server retrieves reviews and stops. The analysis lives in the skill, so pair the connector with the Claude skill or the Codex plugin rather than expecting the server to think.

Claude

Route 1 of 3 · skill + connector

Download the Review Intelligence skill. In Claude Skills, select Add → Upload a skill, choose the ZIP, and turn the skill on.

Add Review Retriever as a custom connector. Use https://reviews.doubledash.me/mcp and leave the optional OAuth client fields empty.

Start a new chat, enable Review Retriever under Connectors, paste the app URL and market, and approve the read-only retrieval tool when Claude asks.

Download the skill →Connect in Claude →

Codex

Route 2 of 3 · plugin

The plugin contains both the Review Intelligence skill and the Review Retriever MCP connection.

codex plugin marketplace add angrysushi11/review-intelligence --ref main
codex plugin add review-intelligence@doubledash

Run the two installation commands in order.

Start a new task so the plugin and its MCP connection load.

Paste the App Store or Google Play URL, add the market when it matters, and ask for the first useful read.

Open the install guide →

Custom GPT

Route 3 of 3 · upload

Nothing to install, and one extra step every time: the GPT cannot fetch reviews itself, so you bring it a file.

Paste an App Store or Google Play URL into Review Retriever.

Export the reviews as Markdown.

Upload the file to the Custom GPT and ask your first question.

Open Review Retriever →Open the Custom GPT →

From Google Play Console

Open Play Console.

Go to Download reports and select Reviews.

Select the app, year, and month.

Download the CSV and paste or upload it.

From App Store Connect

Open App Store Connect and choose Apps.

Select the app.

Open Ratings and Reviews from the sidebar and select the platform.

For a larger export, use the App Store Connect API customer reviews endpoint, then paste the JSON or CSV you create.

Official source controls: Google documents the available review filters and summaries in its Play Console ratings and reviews guidance. Apple documents the corresponding workflow in App Store Connect ratings and reviews. Use those platform views to define the source set before interpreting the themes.

Another MCP client

Add a remote HTTP server named Review Retriever using https://reviews.doubledash.me/mcp. If the client accepts an MCP JSON file, this is the connection used by the Codex plugin:

{
  "mcpServers": {
    "review-intelligence": {
      "type": "http",
      "url": "https://reviews.doubledash.me/mcp"
    }
  }
}

Client menus and configuration filenames vary. No authentication headers or credentials are required by the current public endpoint.

Reference

Configuration

The client normally fills these fields from your request. You only need to name the market or limit when the default is not useful.

url: Required. An App Store or Google Play URL. A numeric Apple app ID or Google Play package ID also works.
platform: auto by default. Use app_store or google_play only when detection needs help.
market: en-US by default. Use a two-letter country or language-country value such as us or de-DE.
limit: 500 by default and at most 500 unique review records per call.
sort: most_recent by default. The other supported values are most_helpful and rating.
include_markdown: false by default. Turn it on only when you also need the legacy Markdown export.

Data boundary

Privacy and data boundary

What you send

The public app URL or identifier, the market, limit, sort order, and whether you asked for a Markdown copy.

What comes back

Public review text and available public metadata, stable evidence IDs, and explicit coverage counts for the returned dataset.

What it does not access

App Store Connect, Google Play Console, private analytics, customer accounts, credentials, or unpublished reviews.

What it changes

Nothing. The tool is read-only, non-destructive, and idempotent. It retrieves public data and returns it to the client.

Storage boundary

The connector does not write retrieved reviews to an application database. The hosted endpoint may still produce ordinary infrastructure request logs, so do not send credentials or private exports through the public MCP server.

Analysis boundary

Reviews support observations and hypotheses. They do not prove churn, conversion, retention, revenue impact, or causality without behavioral or financial data.

First prompt

Analyze reviews for [App Store or Google Play URL] in the US market.
Use Review Retriever, then give me the first useful read.

If the app is country-sensitive, replace “US market” with the storefront you need. Different storefronts can return different review sets.

Connected? Go back to the analysis method.

Back to Review Intelligence →