For AI agents

Your agent writes the tests.
What you do with the extra time is up to you.

Dev Scenario gives Claude Code, Cursor and GitHub Copilot eyes and hands on real Android and iOS devices. Ask for a test in plain words: the agent reads the screen, writes the flow, runs it, reads what failed and fixes it, until it passes.

Two ways in

Use the built-in assistant, or any AI tool you like.

Built into the app

Open the assistant in Dev Scenario and describe what to test, in plain words. It writes the test, runs it on your device until it passes and saves it to your project. It uses your own Claude Code sign-in.

Your own AI tool

Already work in Claude Code, GitHub Copilot CLI, Codex or Gemini CLI? Connect it in one click and ask from there. Any tool that supports MCP works the same way.

The loop

Everything the agent needs to close the loop by itself.

1 · Look

See the screen

The current screen as text the model can read, plus a screenshot, and selectors checked against the live UI before they go into a test.

get_screenfind_elementslist_devices
2 · Write

Write a valid flow

The full command reference, and the real parser to validate a flow before it runs, so no guessing at syntax.

get_command_referencevalidate_flow
3 · Run

Run it on a device

Runs a draft or a saved flow, scenario or chain on the selected device and waits for the result.

run_flowrun_scenariorun_chain
4 · Fix

Understand the failure

The failing step with its screenshot, failed network calls, device logs and the test's history: flaky, newly broken, or broken since a commit.

get_run_reportget_network_logsget_device_logsget_run_history
5 · Keep

Save it where it belongs

The agent reads your project's structure, reuses shared steps like login, and saves the test in the right module, in your naming style.

get_project_structuresave_flowsave_scenario
Beyond writing tests

Ask it for the chores too.

Mock the backend

“Make the payments endpoint return a 500 and check the error screen.” The agent adds mocks and switches mock groups on and off.

Triage a red run

“Why did checkout fail last night?” It reads the report, the failed requests and the logs, and tells you whether it's the app or the test.

Test data and languages

Parameter profiles and translations, so one flow runs for every account and every language.

Deep links

Save deep links in bulk and open them on the device to jump straight to the screen under test.

Git

Status, diffs, commits, branches and merge conflicts in your test project, without leaving the chat.

Android and iOS

Emulators, simulators and real devices, including physical iPhones, with the same commands.

You stay in charge

An agent with real devices, on a short leash.

Stays on your machine

The server is off until you switch it on, listens on localhost only, and every request needs a token. Your app, screens and tests never pass through our servers.

Permission levels

Tools are grouped into read, run, change and delete, and you choose which groups the agent gets. Delete is off until you turn it on.

Ask before changing

By default the agent asks before it changes or deletes anything, and shows a diff of the edit. Allow once, always for that tool, or switch to Auto.

Everything is logged

Every tool call goes to the audit log. Tests are plain .ds text files in your Git repository, so every change is reviewable.

Setup

Connected in about two minutes.

1

Turn on the MCP server

In Dev Scenario, enable the AI Assistant plugin and switch Enable server on.

2

Add it to your agent

The AI Assistant tab connects Claude Code, GitHub Copilot CLI, Codex and Gemini CLI in one click, and shows each one's setup to copy, for example Claude Code:

claude mcp add --transport http --scope user devscenario \
  http://127.0.0.1:8787/mcp \
  --header "Authorization: Bearer <token>"
3

Ask for a test

Connect a device, open your project and ask. Works with any client that speaks the Model Context Protocol.

Let your AI agent test your mobile app.

Free for 6 weeks, no card needed. Running the saved tests in CI is free.