Changelog

Backstage is the new Stunt Double blog: product updates, practical guides, and behind-the-scenes stories from the team, all in one place. Follow along to get more out of your cast and see what we are building next.

A new integrations page covers every way to connect Stunt Double, with a downloadable .mcpb bundle that adds the MCP server to Claude Desktop in a single click. Manual setup for Claude Code and Cursor or Windsurf is a tab away.

Setting up and launching an interview round is quicker. Launching now drops you straight on the Report tab with live progress as sessions complete, and the regenerate control moved up to the page header where it is easy to find.

The sign-in screen is cleaner and quicker to scan, with copy that matches the passwordless code flow rather than the old magic-link wording.

Fixed a bug in GitHub integration authentication so connecting a repository works reliably.

The workspace and project navigation are now organised around three jobs to be done: Discover (interviews and research), Validate (checklists, design reviews, site topology), and Improve (insights, goals, comments). One vocabulary across the sidebar, mobile nav, command bar, breadcrumbs, and the welcome tour.

Every project overview now groups its summary, recent activity, and quick actions under Discover, Validate, and Improve, so the dashboard mirrors the way you actually work, and getting from "what happened?" to the next session takes a single click.

The workspace dashboard surfaces project summaries with the same Discover/Validate/Improve grouping, so you can scan progress across every project without opening each one.

Run Stunt Double browser workers behind your own firewall with a single `docker run`. The image is published as a multi-arch build (linux/amd64 + linux/arm64) on the GitHub Container Registry, so Apple Silicon and Intel hosts both get a native binary.

The recommended setup ships with a Watchtower sidecar that polls every five minutes for new worker images and recreates the container in place. New hosts get auto-update out of the box; existing ones get it the next time you copy the snippet from workspace settings.

Pick a city in your actor settings and every browsing session that actor runs (chat, checklists, insights, topology, interviews) is routed through a Browserbase residential proxy in that location. Useful for testing geo-gated content, localised pricing, and region-specific UX.

Every interview round is now kept on the timeline. Re-run the same discussion guide against a new iteration of your product or Figma file and the previous rounds stay in place, so you can compare findings across versions instead of overwriting them.

Cast members can now carry what they told you in earlier interviews into later sessions and chats. Opt in per interview, and follow-up rounds build on prior context instead of starting from a blank slate.

Design Reviews moved into the Validate section as a tab alongside Checklists and Topology, so all the ways to pressure-test a build live in one place.

Ideas, Goals, and Comments are now one section ("Co-create" inside Improve) with a shared tab bar, so capturing feedback, setting goals, and generating insights feel like one feature rather than three.

The Figma OAuth app is ready for the Figma marketplace and the feature flag is gone, so every workspace can connect Figma and create projects from a file without waiting for an invite.

Prompt caching is now wired through every multi-step Anthropic tool loop (checklists, interviews, insights, chat) with hit-rate instrumentation. Long agent sessions run noticeably faster and cost less per turn.

Web, desktop, and the MCP server now ship under a single version number. When you read about v1.5 in the changelog, you know it matches the build you have installed and the server your agent is talking to.

Recruit a panel from your actors or freshly generated personas, build a discussion guide of questions and think-aloud tasks, and launch a round of moderated interviews. Watch transcripts stream in as each session runs.

Describe an audience and Stunt Double generates a persona, with a name, background, and point of view, to round out your interview panel in seconds.

Point an interview at a Figma prototype and participants react to the design frames, so you can validate direction with no live product required.

Every round wraps with a report that surfaces key themes, verbatim quotes, and recommendations ranked by severity. Findings you can act on, not just transcripts to read.

Re-run the same discussion guide on each new version and compare findings round over round to see what actually improved.

New MCP tools let you create an interview, build its discussion guide, launch a round, and read back the synthesised report straight from your AI client or scripts.

Add an authenticator app as an optional second factor on top of email code and Google sign-in, for an extra layer of account security whenever you want it.

Claim a domain on the AI Agent Readiness Index and it now appears as a project in the app, connecting your public score to the work you do to improve it.

Google Gemini is now scored as a live provider on the AI Agent Readiness Index, alongside Claude and ChatGPT, so comparisons reflect the assistants your customers actually use.

Sign-in, invitation, and notification emails now carry the Marquee look with full dark-mode support, so every message looks right whatever your mail client is set to.

Tidy up your account by archiving workspaces you no longer need. In-flight runs wind down cleanly and the workspace disappears from view, with our team able to recover it if you change your mind.

Project settings now live in the sidebar so they are reachable from any page, and switching workspaces on mobile is faster.

A new public leaderboard at index.stuntdouble.io that scores hundreds of real websites on how well they work for AI agents. Compare providers side-by-side, explore scoring methodology, and see how your domain stacks up.

Domain owners can now sign in, edit metadata, re-run scoring on demand, and manage how their site appears on the Index.

A new visual identity rolls out across the app, marketing site, and desktop shell: refined typography with our new Untitled Sans and Serif fonts, brand icons, a warmer palette, and more considered hover and surface states.

Sign-in, invitations, and notification emails now match the Marquee look, so every touchpoint feels like one product.

A fresh cut of the macOS desktop app shipping the latest product work, including the Marquee refresh, improved project dashboard, and pull-to-refresh on touch devices.

On phones, the Index leaderboard now sits above the stats rail so the rankings are the first thing you see.

Run a design review across any page or prototype and watch multiple actors annotate it from their own perspective. Feedback lands on a shared canvas where you can pan, zoom, resolve comments, and re-run the review after making changes.

Create a project directly from a Figma file. Connect Figma via OAuth and bring your designs into Stunt Double so actors can review screens before you ship a single line of code.

Visualise how the pages of your product connect to each other. Stunt Double crawls your site, including links tucked away in the footer, and builds an interactive map you can explore.

Stunt Double now follows your operating system theme by default, and you can override it any time from your account settings. Dark mode, light mode, or whatever your Mac is doing right now.

The mobile drawer now mirrors the full desktop sidebar, so every navigation option is a tap away when you are working from your phone.

Workspace activity (new feedback, completed checks, fresh insights) now streams in more reliably and with less lag, keeping everyone on the same page without needing to refresh.

Page skeletons now match the layout you are about to see, so navigating between sections feels instant and less jarring.

When you open an insight, its parent project stays expanded and the Insights nav item is highlighted, so you always know where you are.

A signed and notarised native macOS app. Runs as a universal binary on both Apple Silicon and Intel Macs, with automatic OTA updates.

A download button in the navigation always points to the latest release. No manual version tracking needed.

Submit feedback from anywhere in your workspace. The form is always accessible, so your team can share thoughts without leaving what they are working on.

See the status of every piece of feedback you have submitted directly from your account page. Know what has been acknowledged, actioned, or resolved.

Reduced from 32 MCP tools to 23 without losing functionality. Cleaner interfaces, fewer redundant calls, and faster agent interactions.

Creating a checklist from a project now prefills the URL and name automatically, saving setup time.

Create goals directly from insight reports and dynamically link them as new findings come in.

Design system reviewer, security auditor, legal reviewer, and compliance checker join the cast, expanding what actors can specialise in.

Resolved an intermittent issue where an error was displayed during sign-in code entry.

Run browser workers behind your own firewall or VPN. Available on Enterprise plans.

Link your GitHub repositories to workspaces and bring actors into pull request reviews. Stunt Double can now surface user perspective directly in your code review workflow.

Connect your Linear workspace and use Stunt Double as an agent inside Linear. When an issue is assigned to an actor, it gets to work automatically.

Connect Figma via OAuth and let actors interact with your designs before a single line of code is written.

A native Stunt Double desktop app for macOS. Now available for download.

Regenerate AI insight reports after making changes, with a full version history so you can always compare what changed and why.

Assign and configure self-hosted workers directly from the workspace dashboard.

Click anywhere on a live prototype or product to drop a pin, write a comment, and start a thread. Replies, notifications, and screenshot attachments are all included.

Reply to any feedback comment and get notified when someone responds. Keeps conversations in context rather than scattering them across tools.

Comments are pinned to specific page versions using a content hash, so feedback stays anchored even as pages evolve.

The chat assistant can now query your projects, goals, issues, checklists, and feedback in real time, giving answers grounded in what is actually happening in your workspace.

View feedback for a specific page or route, including full support for single-page apps that update the URL without a full reload.

Optimised font loading and asset delivery. The app opens noticeably faster, especially on mobile.

The mobile bottom bar now surfaces project views as primary tabs, making it much faster to move between Goals, Checklists, Feedback, and Workflows on small screens.

AI-generated insight reports synthesise feedback and research into structured findings. Now available to all users as a public beta.

Connect individual feedback items and insight reports to product goals, so you can track how user research supports (or challenges) what you are building towards.

Prominent stats, a two-column layout, and an activity feed give you a clear overview of what is happening across a project at a glance.

The goal detail view now uses collapsible sections and inline feedback search, making it easier to focus on what matters when working through a goal.

See a summary of every project directly from the workspace home, without clicking into each one.

The feedback feature is now available to all users.

Sign in and sign up are now the same step. Enter your email or use Google, and an account is created automatically on first use. No separate signup page needed.

A new split-panel layout on login and signup, with feature highlights on the left and the sign-in form on the right. Cleaner and more confident.

The feedback dashboard now has three goal-oriented views: onboarding for first-time setup, page-grouped review for regular use, and dedicated settings.

Gather feedback from each session in one place, attached to the project it came from, so the signal stays close to the work it is about.

A first Model Context Protocol server lets you drive actors, projects, and checklists from your own AI client or scripts.

Get notified when a long-running session finishes or when there is something waiting on you, so you do not have to keep the tab open.

Actors run inside a hosted browser session and act on live pages using natural-language instructions, so they can work through real products rather than just talk about them.

Every step is grounded in a screenshot of the current page, so an actor sees what you would see before it clicks, types, or navigates.

Actors chain actions across a session, reading the result of each step before deciding the next, which makes longer tasks far more reliable.

Describe what should be true about a page or flow and let an actor verify it, so routine checks happen without you driving every click.

Point a checklist at any address and kick off a run, with the URL and name prefilled from the project where it lives.

Each run reports every step with its outcome and supporting detail, so you can see exactly what passed, what failed, and why.

A workspace gives your team one shared home for actors and projects, with members and settings in a single place.

Group related work into projects so each product or initiative has its own space for actors, sessions, and results.

Invite teammates into a workspace by email and start collaborating on the same actors and projects right away.

Spin up an AI actor with its own name, background, and point of view, the foundation everything else in Stunt Double is built on.

Hold a back-and-forth conversation with an actor to pressure-test ideas and hear a perspective that is not your own.

Stunt Double opens to a small group of early users to shape where the product goes next.