Skip to main content

Stunt Double + Figma

Figma is where the design decisions happen; Stunt Double is how you find out whether users can actually follow them. The integration works at two levels: agent-driven testing of prototypes (via MCP), and event-driven automation that re-checks work whenever a file changes (via the Figma connection).

Who this is for: designers validating flows before handoff, design systems teams guarding consistency, and PMs who want usability signal attached to the file, not in a separate report.

Setup

  1. Connect Figma to your workspace: Settings -> Connections in app.stuntdouble.io, connect Figma. This enables Figma event triggers for workflows (file updates, comments, version updates, library publishes).
  2. Link files to projects: set the Figma file URL on a project so runs, feedback, and interviews stay associated with the design source.
  3. For agent workflows, add the Stunt Double MCP connector to whichever agent you use alongside Figma (Claude, Claude Code, Cursor): https://app.stuntdouble.io/api/mcp. If you also run Figma's own MCP server, the two compose well: Figma MCP reads and edits the design, Stunt Double MCP tests it.

Workflow: test a prototype with personas

Actors drive real browsers, so anything viewable at a URL is testable. For Figma that means:

  • Prototype share links (figma.com/proto/...) set to "anyone with the link can view". Good for click-through flow testing: can a persona find their way from the landing frame to the goal?
  • Figma Make published sites for anything with real interaction. See the dedicated Figma Make guide.

From your agent:

"Use the validate_design prompt on my prototype link with personas for our two main segments. Focus on whether the new onboarding order makes sense."

The agent creates an interview, adds task items ("Get from the welcome screen to creating your first project"), attaches varied personas (mix devices; prototypes often fail on mobile first), launches, and returns the synthesised report with transcripts as evidence.

Interpreting results on prototypes: personas can only follow the interactions you wired. A "could not proceed" finding on a prototype sometimes means a missing hotspot rather than a design flaw; the transcript makes it obvious which.

Workflow: re-check automatically when the design changes

This is the predictive half. Once Figma is connected, workflows can trigger on Figma events:

  1. Create checklists for the flows the design feeds (staging or production URLs).
  2. Create a workflow with a Figma event trigger and steps that run those checklists, plus a notification step.
  3. Activate it.

Now a designer shipping a file update or publishing a library triggers a re-check of the affected product flows, and discrepancies between the design intent and the built product surface within minutes of the change, not at the next QA pass. An agent can set this whole thing up from the setup_guardrails prompt.

Workflow: brand and design-system guardianship

Give an actor your brand guidelines as knowledge (voice, spacing rules, colour usage, component dos and don'ts) and use it to run checklists like:

  • "Every page uses the product font stack; no default serif fallbacks anywhere."
  • "Primary CTAs use the brand primary colour and sentence case."
  • "Empty states show the illustration style, not stock icons."

Run these against staging on a schedule or on library publishes, and the design system stops drifting silently. The check_brand and check_design_system MCP prompts encode this whole workflow; the validation guide covers the pattern in depth, including compliance and cross-surface continuity audits.

Tips

  • Keep prototype links public-viewable for the duration of a test; personas do not have Figma accounts.
  • Name interviews after the design iteration ("Onboarding v3, post-crit") so before/after comparisons are easy.
  • Feedback captured in Stunt Double carries page URL, pin position, and screenshots, which maps naturally back to frames when you triage with the triage_feedback prompt.