Agentic QA Connect

Install and launch the app

Download the latest build for your OS from the releases page, install it like any other desktop app, and launch it.

Windows

Download Agentic-QA---connect-Setup-<version>.exe and run it.

Windows: allow the installer to run

The installer isn't code-signed yet, so Windows SmartScreen is likely to stop it with "Windows protected your PC". Click More info, then Run anyway.

Windows: run the installer

1. Choose Installation Options — pick Only for me. It needs no administrator rights, while Anyone who uses this computer installs into Program Files and asks for them.

   

2. Choose Install Location — keep the suggested folder or pick another one, then click Install.

   

3. When setup is complete, leave Run Agentic QA - connect ticked and click Finish.

   

macOS

Download the .dmg for your Mac — the one ending in -arm64.dmg for Apple silicon, the other one for Intel — and open it.

macOS: allow the app to run

The app isn't signed with an Apple Developer ID yet, so the first time you open it macOS refuses with "“Agentic QA - connect” is damaged and can't be opened". The app isn't damaged: that's how macOS reports an unsigned app downloaded from the internet. Click Cancel.

Then let macOS open it:

1. Open Terminal (Applications → Utilities → Terminal).
2. Paste this command and press Return. It removes the "downloaded from the internet" mark from the app:

   xattr -dr com.apple.quarantine "/Applications/Agentic QA - connect.app"

3. Open the app from Applications again. From now on it starts normally; run the command again after you install a new version.

On an Intel Mac, macOS may say instead that it can't verify the app; the command fixes that too. Run it only for the app you downloaded from the releases page.

Connect it to your installation

Everything happens on one page of the installation: open the account menu (your avatar, top right) → User Settings → the Agentic QA Connect tab. Every user has it; the tab appears once Agentic QA Connect is enabled on the installation. Connect the machine automatically with one click, or manually when the one-click link can't reach the app.

Automatic Manual

Recommended: one click in the installation, nothing to copy by hand.

1. Make sure the app is installed and running on the machine you're sitting at, and open the installation in a browser on that same machine.
2. Click Connect app next to the Deployment URL. The browser asks whether it may hand the agentic-qa-connect:// link to the app — confirm, and the app comes to the foreground.

   

   

3. The app asks "Connect this machine to <your installation>?". Click Connect. Nothing is applied until you confirm — a deep link can be triggered by any web page, so the app never reconfigures itself silently.

   

   

That's it: the app fills in the Deployment URL, trades the one-time code from the link for its own handshake token, saves the configuration and starts connecting. Nothing is copied by hand.

The one-time code is valid for 60 seconds and works once. If you take longer, the app says "The link has expired — click 'Connect app' again." — go back and click the button again.

Use this when the app runs on a different machine than the browser, or the one-click link doesn't reach it.

1. Copy the Deployment URL from the Agentic QA Connect tab in User Settings and paste it into the app under Config → Connection → Deployment URL.

   

   

2. Copy the Client ID from the top of the app's Dashboard, paste it into Pair by Client ID on the same tab and click Pair.

   

   

3. A new row appears under Paired clients with its handshake token already issued — click Copy.

   

4. Paste it into the app under Config → Connection → Auth Token and click Save.

   

5. Back on the Dashboard, click Start.

About the handshake token

The token is what the installation checks when the app connects — without a valid one the connection is refused, no matter what the Deployment URL says. A few things worth knowing:

• It is personal and per-machine, and it expires after one month; the paired row shows how long is left.
• Regenerate issues a new one and kills the old one immediately — paste the new value into the app.
• Revoke stops this machine connecting until you issue a new token.
• A machine belongs to exactly one user, so a handed-over laptop always has a single owner and a single live token. To take over a machine someone else paired, use Connect app on that machine: pairing by Client ID is refused while the machine belongs to another user.
• Deleting a user's account removes their pairings and tokens: the machines they paired stop connecting until someone pairs them again.

Check that it is online

In the app, the status bar flips to Connected and shows the Deployment URL it reached. On the installation, the row under Paired clients shows a green ● Online badge. The badge is driven by a heartbeat, so a machine that dies without disconnecting cleanly turns ○ Offline after about ten minutes.

Display Name (Config → Connection) is what the installation labels your machine with in the client dropdowns; leave it empty and the app sends the machine's hostname.

Set up the browser

A fresh install already carries a built-in Playwright_MCP server — that's the one Local Machine web sessions drive, and you configure it from Config → Playwright MCP (browser, capabilities, headless, emulated device, viewport). Use the Install button there once per browser: Playwright needs its own browser binaries.

Run a web session

In the test-session form, under Where to Test, pick Local Machine as the Browser. The Connected client dropdown then lists your paired and currently-online machines.

Add your MCP servers

Your own servers live on the MCP Servers tab of the app. Each card is one server:

• Name — what the server is called in the installation's pickers. Only letters, digits and _ survive; anything else is rewritten.
• Type — stdio (a local process), http or sse (a URL).
• Command + Args for stdio; URL + Headers (JSON) for the other two.
• Stateful — keeps one live MCP connection per test session instead of reconnecting per call.
• Test — connects right there and reports how many tools it found, before you involve the installation.

Edit JSON shows the same thing as a file. The shape is:

{
  "deployment_url": "https://connect-gateway.3873d9e0-d8bf-41a0-9c2e-ae14f02ae4fc.testinator.ai",
  "auth_token": "…",
  "timeout": 120,
  "ssl_verify": false,
  "display_name": "DESKTOP-MA6W47A",
  "servers": {
    "Playwright_MCP": {
      "type": "stdio",
      "command": "npx",
      "args": ["@playwright/mcp@latest", "--caps", "vision", "--isolated"],
      "stateful": true
    },
    "my_api": {
      "type": "http",
      "url": "http://localhost:8787/mcp",
      "headers": { "Authorization": "Bearer MY_TOKEN" },
      "stateful": true
    },
    "legacy_thing": {
      "type": "sse",
      "url": "http://localhost:9000/sse"
    }
  }
}

Tokens, paths and process commands never leave your machine — the installation only ever sees the tool calls and the list of tools each server advertises.

Run a Desktop App session

To test the MCP servers themselves, create a test session for a product of type Desktop App. There is no "What to Test" URL list here — Where to Test covers both the machine and what runs on it.

No such product yet? Create one under Admin → Products → New with Product Type set to Desktop App. The type is listed only when Agentic QA Connect is enabled on the installation.

1. Pick the Connected client.
2. Pick one or more MCP server cards. Each card shows how many tools and resources that server advertises.
3. Narrow down the Tools and Resources chips if you don't need all of them — the form warns you if you turn every tool off, because the agent then has nothing to work with.

Prepare the machine

Test a native Windows application on your own Windows machine: the agent sees the screen and drives the mouse and keyboard through the windows_mcp server in the app.

• Windows 10 or 11, x64, with the app installed and connected — see Installation.
• A logged-in, unlocked desktop for the whole run. A locked screen or a disconnected Remote Desktop session stops every click and keystroke — turn off the screen saver and the automatic lock.
• Don't use the mouse or keyboard while checks run. The agent shares them with you, and anything you click moves the application out from under it.
• The application under test shouldn't run as administrator: Windows doesn't let a regular process control elevated windows. If it has to run elevated, start the Agentic QA Connect app as administrator as well.
• Set the Windows display language to English.
• Run one session per machine at a time — two sessions would fight over the same screen.

Add the windows_mcp server

In the app, open MCP Servers → + Add Server, fill it in exactly like this and click Save:

• Use exactly this name. The installation finds the driver by it: a server called Windows_MCP isn't recognised, and the session form finds no driver on the machine.
• Keep Stateful on. Without it every single click starts the server from scratch.
• Keep the exact version in Args. With @latest the server looks for a new release on every start and downloads it when there is one, so a session can fail to start at a random moment.
• Nothing else to install: uvx runs through the copy of uv that ships with the app.

Test the server

Click Test on the windows_mcp card. A working server reports how many tools it found. Then click Stop and Start on the Dashboard so the installation sees it.

Create a Desktop App Windows product

In the installation, open Admin → Products → New, name the product and set Product Type to Desktop App Windows. The type is listed only when Agentic QA Connect is enabled on the installation.

Run a Windows session

Create a test session for that product — the Windows MCP Test Workflow is picked for you. Under Where to Test:

1. Connected client — your machine.
2. Driver — windows_mcp. It's offered only when the selected machine's app has a server with that name.
3. App to launch (optional) — the application's name as it appears in the Start menu; it's launched at the start of every check. Leave it empty to run the checks against whatever is already open, for example an application you started by hand or a Remote Desktop window.

Setup Notes are optional too: the agent gets them with every check, so use them for anything it should know about the machine or the application.