Start local test authoring from wherever you work: the mabl agent chat, your terminal, or an AI client like Claude or Cursor. For apps running on localhost or behind a VPN, you can now have the mabl agent build tests on your machine without jumping between separate tools.
From the mabl agent chat
Ask the mabl agent to build the test on your machine. The agent sets up the authoring task, and the task waits for you to start it. You can click Start in desktop app, or copy the mabl CLI command shown beside the button and run it in your terminal.
The same two options appear on the task's Work tab and in the tasks table, so you can start the task later from either place.
From the mabl CLI
With mabl CLI version 2.144.10 or later, start a waiting local authoring task by its session ID:
mabl agent authoring start --session-id {session-id}
The command works for any local task, whether you set it up in the web app, in the mabl agent chat, or from your AI client. You don't need the mabl Desktop App to build the test.
From your AI client
The mabl_authoring_initiate_local tool in the mabl MCP now returns the mabl CLI command that starts the task, instead of a link that opens the Desktop App. Your AI client can give you the command to run, or run it in your terminal itself. For example:
Have the mabl agent build a test on my machine for the checkout page at http://localhost:3000.
Also new in the mabl agent chat
- Fix the setup of authoring work. When an authoring task the agent started gets stuck on its setup, such as a missing login credential, the agent can correct the task's credentials, application, environment, URL, DataTable, or scenario and start it again.
- Run data-driven tests with a scenario. Ask something like "Run the shipping test with the Cambridge scenario," and the agent runs the test once for each scenario you name.
- Look up test cases by ID. Ask about a TestRail or Zephyr test case, an Xray test, or a Jira or Azure Boards issue by its ID, and the agent finds it in whichever connected integration holds it. If more than one integration has a match, the agent asks which one you mean.
- Agent instructions apply in chat. The agent follows the agent instructions your workspace has written for test authoring and results analysis.