Skip to main content

Stunt Double + Figma Make

Figma Make turns a prompt or a design into a working web app and publishes it to a real URL. That published URL is exactly what Stunt Double actors are built to test, which makes Make plus Stunt Double the fastest idea-to-evidence loop available: describe a product, publish it, and have AI personas try to use it, all before an engineer touches it.

Who this is for: designers and PMs prototyping in Figma Make, and founders validating concepts before committing to a build.

Setup

  • Add the Stunt Double MCP connector to the agent you use around Make (Claude, Claude Code, Cursor): https://app.stuntdouble.io/api/mcp. Sign in when prompted the first time.
  • Optionally connect Figma in Settings -> Connections at app.stuntdouble.io so workflows can also react to Figma events.

The loop: build, publish, test, revise

  1. Build in Make. Get the prototype to the point where the core flow exists, even roughly. Real (fake) content beats lorem ipsum: personas read the page like users do.

  2. Publish. Use Make's publish action to get the live site URL. Stunt Double needs nothing else; there is no SDK to embed.

  3. Test with personas. From your agent, use the validate_design prompt with the published URL, or be specific:

    "Interview 4 personas against my published Make site: a power user of a competitor, a non-technical small business owner, a price-sensitive student on mobile, and an accessibility-conscious user. Brief: can they understand what this does and complete signup?"

    The agent builds the interview (tasks plus comprehension questions), attaches the personas with mixed devices, launches, and returns the report.

  4. Revise in Make. Take the top friction points back into Make (or let your agent rewrite the Make prompt with the findings folded in), republish, and re-run the same interview. Because the discussion guide and personas are unchanged, differences between rounds are attributable to the design change.

Two or three cycles of this typically settles the big questions (comprehension, navigation, willingness to proceed) in an afternoon.

When the prototype graduates

Make prototypes tend to become real products. The assets you built during validation come along:

  • The project in Stunt Double just gets a new URL (staging, then production).
  • The interviews re-run against the real product, giving continuity of evidence from prototype to launch.
  • Add checklists for the flows as they harden, and a scheduled or deploy-triggered workflow (setup_guardrails prompt) so the validated behaviour stays validated. See the codebases guide for the engineering-side loop.

Tips

  • Test the ugly version. The strongest signal comes before you polish; personas are immune to embarrassment.
  • One flow per round. "Can they sign up" and "do they understand pricing" are separate studies; mixing them muddies the report.
  • Watch a session recording in the dashboard at least once per round; seeing a persona hesitate is more persuasive to stakeholders than any summary.
  • Make sites are public by default when published, so there are no auth hurdles. If you add auth later, actors handle email flows: they have their own inboxes for OTP and magic-link sign-ins.