AI agents

Let an MCP-compatible AI app, such as Claude Code or Claude Desktop, record, run and fix tests in Actify Studio — while you watch every step.

An AI app on your PC that supports the Model Context Protocol (MCP) — for example Claude Code or Claude Desktop — can use Actify Studio like a QA engineer sitting next to you: it reads your tests and runs, uses the app’s browser, records and edits tests, runs them, explains why they fail and writes bug reports. Every action happens in the app, in front of you.

AI agents are off until you turn them on. Actify Studio has no AI model of its own and sends nothing anywhere: what the agent reads is handled by the AI app you connect, under its own terms.

Connect an AI app

Open Settings → AI agents and turn on Let AI agents use Actify Studio. The status says when the app is waiting and which client is connected.
Under Connect an AI app, follow the way for your app:
- Claude Desktop — click Get the extension: the app creates actify-studio.mcpb and shows it in its folder. Drag it into Claude Desktop → Settings → Extensions and choose Install. No Node.js needed, and the extension stays up to date with the app.
- Claude Code (terminal) — copy the command shown and run it once (Node.js required).
- Other MCP apps (configuration file) — copy the JSON shown into the app’s MCP configuration (Node.js required).
Ask the agent something, e.g. “Record a checkout test with an expired coupon, then run the smoke tag.”

For Playwright users

Terminal

claude mcp add actify-studio -- node "%USERPROFILE%\.actify-studio\actify-mcp.cjs"

TypeScript

{
  "mcpServers": {
    "actify-studio": {
      "command": "node",
      "args": ["C:\\Users\\<you>\\.actify-studio\\actify-mcp.cjs"]
    }
  }
}

Copy the exact path from Settings. The server is a small Node script that talks to the running app through a local, per-user channel; it works only while Actify Studio is open with the option on.

While an agent is connected

The AI agent connected bar under the toolbar shows the client, what it is doing (Reading the tests…), the number of requests, Activity, the gear for its settings and Disconnect, which cuts it off at once.
Steps it records appear in the Action History as it works, like yours.
When a choice is yours, it asks in the app — buttons or a text field — and the bar shows Waiting for your answer.
Deleting tests, suites or unsaved steps, replacing a saved test, or stopping a run you started always asks for your confirmation in the app.
One agent at a time. If the Actify Studio window is minimized the agent can’t act on the page, and says so.

What the agent can do

Area	What it can do
Read	Projects, tests with their steps, Pages, Reusable steps, run history with errors and the failed step, the page’s network calls and console (the latest 50 and 100, even when not recording), the texts on the page.
Run	A test, a folder, a suite, a tag or the whole project, Headless or In-App; wait for the end; see why a test failed, with the screenshot of that moment. Stop the runs it started.
Browse and record	Open pages, click, type, choose options, press keys and take screenshots — on a phone or tablet too. Record tests with checks and variables, then save them.
Edit	Open a test and change, move, delete or insert steps; try steps on the page; set test data; rename, duplicate, move and tag tests; create and change suites; switch environment; set the device.
Pages and healing	Add elements to Pages, see where an element is used, fix a Page element (you choose which tests to update), use a healed locator.
API and mocks	Read, send and import (from cURL) API requests, turn a network call into an API request, add API steps and mocks to a test.
Quality	Check accessibility (WCAG 2 A/AA), add a Page health check, see what the tests cover, write and update bug reports.

The AI app also offers ready-made prompts: record a test from a description, plan the tests of a feature, explore an area and report bugs, investigate failed tests, run a tag and report, review a test, triage flaky tests.

It works like a QA engineer

It follows a testing playbook: negative and boundary cases, good checks and locators, what to do when a test fails, flaky tests, bug reports.
It reads your project’s QA guidelines & app knowledge (see Projects) to design tests and before calling a failure a bug. Describe your team’s rules and how your app works there.
When it saves a test, it gets notes on what is wrong — no checks, fixed waits, fragile locators.
To change or remove a check of a saved test it must give a reason (the test was wrong, the app changed on purpose, you asked), so an app bug is reported, not “fixed” by weakening the test.
Text on the page and in network responses is treated as data, never as instructions.

Permissions

In Settings → AI agents → Permissions choose what an agent may do. The change applies from its next request; an action not allowed is refused with the reason, and below Full the agent bar shows the level.

Level	The agent can
Read only	Read the project, tests, results, the page and its network, and ask you questions. It changes nothing.
Test	Read only, plus: browse and act on the page, run tests, try steps, send API requests, change the device, write bug reports. Tests stay as they are.
Edit	Test, plus: record, change and save tests, steps, suites, tags, Pages, API requests, mocks and the environment.
Full (default)	Edit, plus: delete tests and suites and replace a saved test — you still confirm each one.

Activity log

Everything an agent changes or runs is written to the Activity log — open it from the agent bar or Settings → AI agents, with a filter. It is stored as agent-log.jsonl in the project (the latest 2,000 entries) and is not exported.

Something unclear or missing? Write to info@actify-studio.com.

Connect an AI app#

While an agent is connected#

What the agent can do#

It works like a QA engineer#

Permissions#

Activity log#