Check Your Google Play Closed Test with an AI Agent (MCP)

During a Google Play closed test, developers keep coming back to the same few questions: Is my test running normally? Did the version I just released reach the test devices? Is the app crashing? Doply runs an MCP server (beta) so you can ask an AI agent those questions in chat instead of opening the dashboard.

This guide explains what MCP is in a nutshell, how to connect the Doply MCP server to Claude Code, Claude Desktop, and Cursor, and what you can ask once it is connected.

Beta notice: The Doply MCP server is a beta feature. Answers may be inaccurate and features may change. Please let us know if anything looks wrong.

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI tools pull data from external services. You register a server URL and a key in an MCP client such as Claude or Cursor once, and from then on the AI can fetch what it needs from that server in the middle of a conversation. With the Doply MCP server connected, asking "How many days are left in my test?" makes the AI look up your app's actual test records on Doply and answer from them.

Why it helps during a closed test

Personal Google Play developer accounts must run a closed test with at least 12 testers for at least 14 days before applying for production access. Doply runs that test for you, and while it is running, the questions developers ask are fairly predictable:

• Is the test going well? How many days are left?
• I released a new version yesterday. Is it installed on the test devices?
• Has the app crashed on any device?
• What does the app actually look like when it launches?
• What should I write in the production access survey?

All of this is available in the Doply dashboard, but if you already build with an AI tool, asking without leaving your editor or chat is faster. In a tool that also sees your code, such as Claude Code or Cursor, you can look at a crash and then track down and fix the offending code in the same conversation.

What you can check with Doply MCP

With your MCP key, you can look up the following for your own apps.

Without a key, you can still use FAQ search, requirements, service info, and contact details. That is handy if you are not a customer yet and want to ask something like "What do I need to use Doply?"

Step 1: Get your MCP key

1. Log in to Doply.

2. Open AI Agent (MCP) in the dashboard sidebar, or go to /en/dashboard/mcp.

3. Copy your key from Your MCP key. It starts with dply_u_.

The same page shows the connection command and config with your key already filled in, so you can copy and paste them as is.

Treat the key like a password and do not share it. If it leaks, regenerate it on the MCP key page. The old key stops working immediately, so update the key in your AI tool's settings as well.

Step 2: Connect your AI tool

Here are the Doply MCP server details.

Keep the www in the URL. If you use the bare doply.io domain, it redirects to www, and clients drop the Authorization header along the way, so your key never arrives.

In the examples below, dply_u_... is a placeholder. Replace it with your own MCP key.

Claude Code

Run this in your terminal:

claude mcp add --transport http doply https://www.doply.io/mcp --header "Authorization: Bearer dply_u_..."

Then restart Claude Code and ask something like "Show me the status of my apps in testing."

Cursor and other clients that use mcp.json

Cursor registers MCP servers in an mcp.json file. Add the following to it. If you already have other servers, just add the doply entry inside mcpServers.

{
  "mcpServers": {
    "doply": {
      "type": "http",
      "url": "https://www.doply.io/mcp",
      "headers": {
        "Authorization": "Bearer dply_u_..."
      }
    }
  }
}

Claude Desktop and other MCP clients

Each client registers remote MCP servers a little differently. Whatever the client, you need two values: the server URL (https://www.doply.io/mcp) and the Authorization header (Bearer plus your key). Enter them wherever your client configures remote MCP servers. For a plain-text overview of Doply that AI tools can read, see llms.txt.

Step 3: Ask away

Once connected, just ask in plain language. The questions below are the ones shown on the Doply dashboard, and the answers are illustrative examples. Real answers depend on your app's test records and your AI tool.

You: Is my test going well?

AI (example): Yes, your test is running and today is day 9. The app launched on all 16 devices over the last two days, and there are 5 days left until Google's 14-day requirement.

You: Show me the status of my apps in testing.

AI (example): You have 2 apps in testing. Your notes app is on day 9 and installed on 16 devices. Your puzzle game is on day 3 and installed on 15 devices.

You: When does today's run happen?

AI (example): Doply checks for a new version and updates the devices every day at 00:00 KST, then launches apps in sequence from 02:00 KST. The exact time for each app varies a little from day to day.

You: I uploaded a new version. Has it reached the devices?

AI (example): 12 of the 16 devices are on 1.0.3, and 4 are still on 1.0.2. Doply checks for new versions every day at 00:00 KST and updates the devices.

You: Any crashes?

AI (example): Yes. On the current version, 1.0.3, the app crashes right after launch on all 16 devices. The cause is a NullPointerException in MainActivity.onCreate. While the app is crashing, the closed test is not proceeding normally, so please release a fixed version quickly.

You: Show me the latest screenshots.

AI (example): Here are the 16 most recent launch screenshots. Device 3 shows a black screen. Large apps and games may not have drawn their first screen yet 2 seconds after launch.

You can also ask things like "How many updates have been done?", "What are my production survey answers?", or "What do I need to use Doply?"

Reading the answers

New versions reach the devices the next day

Every day at 00:00 KST, Doply checks for a new version and updates the test devices, and from 02:00 KST it launches apps in sequence. Because many apps are launched one after another, each app's exact launch time shifts a little every day. So if you release a new version today, it reaches the test devices the next day. If you ask "Did it reach the devices?" right after releasing, seeing the previous version is expected.

If you see a crash, ship a fix quickly

If the app crashes, the closed test does not proceed normally. Release a fixed version as soon as you can. It reaches the devices the next day, and if later launches show no crash, the issue is resolved. A crash with the status resolved means it only happened on older versions and the current version runs fine.

Launch results (ok, crashed, not running) have been recorded since October 25, 2026, so earlier dates may have no results.

When a screenshot is black

Launch screenshots are taken 2 seconds after the app starts. Large apps and games may not have drawn anything yet at that moment, so the screenshot can come out black. Rather than treating a black screenshot as a crash on its own, check it against the crash results.

FAQ

Can I connect without a key?

Yes. Without a key you can only use FAQ search, requirements, service info, and contact details. To see your own app's test status, you need an MCP key.

Which AI tools does it work with?

Claude Code, Claude Desktop, Cursor, and any other MCP client that supports remote MCP servers over Streamable HTTP with an Authorization header.

I connected, but my apps do not show up.

First check that the server URL includes www: https://www.doply.io/mcp. Without www, the header is dropped and the connection behaves as if there were no key. If you regenerated your key, update it in your AI tool's settings too.

Does checking with MCP help me get production approval?

MCP is simply an easier way to check your test status. Google decides whether to grant production access. That said, it can help you catch crashes early and release a fix sooner.

Run your closed test with Doply

Doply runs your Google Play closed test for you. Google requires 12 testers; Doply provides 14 to 16 testers and keeps the test running for 18 days, a buffer on top of Google's 14. Testing happens on 36 real Samsung Galaxy devices (Android 13/14), starts right after payment, and runs 24/7. It costs $6.99 per app. More than 4,000 apps have passed closed testing with Doply, and more than 3,000 have been approved for production. Google decides on production access, so Doply cannot guarantee approval.

To follow your test from your AI tool, get your key on the AI Agent (MCP) page and connect. For the full closed testing process, see the complete guide to Google Play closed testing.