Create and launch a test with the UserTesting MCP (Open beta)

On this page:

• What you can create
• Before you create a test
• Create a test draft
• Choose your participants
• Launch a test
• When a test stops at a draft
• If your workspace requires approval to launch
• Find your existing tests
• Read the AI summary for a completed test
• FAQ
• Related content

What you can create

The UserTesting MCP creates three test types. Your AI client picks the type from your description, or you can name it.

Live Conversation is not supported. You cannot create, schedule, or launch a Live Conversation test through the UserTesting MCP, because it is scheduling-based. Create it in UserTesting instead.

A test you create through the UserTesting MCP can include instructions, spoken and written responses, rating scales, multiple choice, rank order, matrix questions, Net Promoter Score questions, images, website navigation tasks, and Figma prototype tasks. Tests run on a computer by default; ask for a mobile test if you want participants on a phone.

Before you create a test

Creating a test through the UserTesting MCP requires two things beyond a working connection. See Connect the UserTesting MCP to your AI client (Open beta) if you have not connected yet.

Your account must not have opted out of Generative AI features. If it has, your AI client returns No Accounts with Generative AI and MCP access enabled and cannot list your accounts. Ask your UserTesting account admin to enable Generative AI features.

You need WRITE or ADMIN permission for that test type in that workspace. Permissions are held separately for think-out-loud tests, surveys, and interaction tests, so you may be able to create one type and not another in the same workspace. With READ permission, you can find tests and read AI summaries, but attempting to create one returns a message that you do not have create permission — you can still share the proposed test with an admin.

Create a test draft

The UserTesting MCP always creates a test as a draft first. Creating is never launching. To go live, you take a separate step - see Launch a test.

To create a test draft:

1. Ask your AI client to create a UserTesting test and describe what you want to learn. Describe it the way you would brief a participant taking the test, not as step-by-step instructions for software. For example: You're shopping for a desk. Starting from this site, find one you like and add it to your cart.
2. Select the UserTesting account you want to work in when the client lists your accounts.
3. Select the workspace. Only workspaces where you have permission to create that test type are valid.
4. Decide how you will recruit participants — see Choose your participants.
5. Review the tasks and questions the client proposes, and ask for changes before it creates anything.
6. Confirm. The client creates the draft in UserTesting and returns a link to the test builder.

Leave personal and sensitive details out of your description. If the description is too vague to turn into participant tasks, the client asks you for a clearer one rather than guessing.

Tip: Your AI client asks which account and workspace to use each time. To stop the re-prompting, save those preferences in the client's memory or configuration file — for example, a CLAUDE.md file for Claude Code.

Open the test builder link any time to review or edit the draft in UserTesting before it goes live.

Choose your participants

The UserTesting MCP offers two ways to recruit for a test. You choose one when you create the draft.

Panel audience. Your AI client lists the saved audiences available in the workspace, and you select one. Only auto-sourced panel audiences are returned. The UserTesting MCP cannot build an audience or write screener questions for you, so create the audience in UserTesting first and give it a clear name — you select it by name during the prompt.

Invite link. No saved audience is needed. Tell the client how many participants you expect, and it creates a dedicated audience for that test and returns a participant share link. Send that link to the people you want to take the test.

Launch a test

Launching makes a draft live and starts recruiting participants. Two gates are checked every time you launch, and both are enforced by UserTesting rather than by your AI client.

To launch a test:

1. Ask your AI client to launch the test. The word launch means go live and start recruiting. Say create a draft instead if you want it to stop before going live.
2. If the launch succeeds, the client returns a results link for the running test.

Note: A launch through the UserTesting MCP consumes participant sessions exactly as a launch from UserTesting does. Your saved audiences are a guardrail — they cap how many participants a launch can recruit — so keep audience quotas purposeful.

When a test stops at a draft

Three things stop a test at a draft. In each case, the draft is created, and you finish it in UserTesting.

An image task. Images must be stored in UserTesting, and the UserTesting MCP has no upload path. It creates the draft with a placeholder and returns the test builder link. Upload the image there, then launch from UserTesting.

A Figma prototype task. The success path for a Figma task is defined across Figma components, which cannot be set from a chat window. The UserTesting MCP sets the prototype's starting frame as a placeholder so the draft validates; you set the real success path in the test builder before launching.

Tip: To launch a prototype test end to end without touching UserTesting, use a navigation task with a start URL and a success URL instead of a Figma task. 

If your workspace requires approval to launch

Some workspaces require a reviewer to approve a member before their tests can go live. This is a workspace setting, not a limitation of the UserTesting MCP, and it applies whether you launch from UserTesting or from your AI client.

When approval is outstanding, the UserTesting MCP creates the draft, does not launch it, and returns a link so you can track the test. A workspace admin updates your approval status in UserTesting.

You can ask your AI client to check your approval status at any time without attempting a launch. Approval is a manual admin action that can take hours, so check occasionally rather than repeatedly. Once you are approved, ask the client to launch the test again.

Find your existing tests

You can search and list tests in a workspace through the UserTesting MCP without creating anything. You see only tests whose type you have READ permission for.

• Search by title. Quoted phrases, exclusions, and OR all work — for example, find my UserTesting tests about "checkout flow".
• Filter to your own tests. Say my tests or tests I created. Without that, the client returns every test in the workspace.
• List with filters. Filter by test type or by a date range.

A listing returns 10 tests at a time by default and a maximum of 50. Ask for the next page to see more.

Read the AI summary for a completed test

The UserTesting MCP returns the test-level AI summary for a launched test. Individual session transcripts, task-level responses, and quantitative data are not available through the UserTesting MCP; use the UserTesting API for those.

To read the AI summary:

1. Wait until participants have completed the test. The summary is generated from session transcripts, so it does not exist before then.
2. Ask your AI client for the AI summary of that test.

The summary includes the research goal, an overview, the participant count, timestamps, and a list of key findings. Each finding has a title, a short summary, and a label:

Need · Pain point · Opportunity · Expectation · Process · Ease of use · Language · Trust · Satisfaction · Suggestion · Confusion

Every finding also links back to the specific task in your UserTesting results that produced it, so you can check a finding against the underlying session data. Open those links before acting on a finding.

If the summary is not ready yet, the client reports AI summary not ready for that test and tells you to try again once participants have completed it. Test-level AI summaries cannot be edited or regenerated through the UserTesting MCP.

FAQ

Will it launch a test by accident?

Only if you tell it to launch. The word launch means go live and start recruiting; create a draft stops at a draft. Your saved audiences are a second guardrail, because they cap how many participants and how much spend a launch can trigger.

Why did my test stop at a draft instead of launching?

An image task, a Figma prototype task, or an outstanding workspace approval. See When a test stops at a draft.

Do I always need a saved audience?

No. A panel test requires one. An invite-link test does not — you state the number of participants and get a share link back.

Which test types work through the UserTesting MCP?

Think-out-loud tests, interaction tests, and surveys in the New Experience. Live Conversation is not supported.

Can I create a Live Conversation through the UserTesting MCP?

No. Create it in UserTesting — see Live Conversation: New experience (beta).

Does it ask for my account and workspace every time?

Yes, unless you save those preferences in your AI client's memory or configuration file, such as a CLAUDE.md file.

Can I get session transcripts or quantitative data?

Not yet through the UserTesting MCP. Both are available today through the UserTesting API.

Can I edit the AI summary?

No. Summaries cannot be edited or regenerated through the UserTesting MCP.

Who do I contact about a problem?

Email support@usertesting.com.

Related content