Integrations
A design gets updated in Figma, a pull request opens on GitHub, a preview lands on Vercel. Each of those is a moment when someone should try the flow before it ships. Integrations let Stunt Double act at that moment instead of waiting for someone to remember to start a run.
Actors drive real browsers through real products and report back with evidence. Anything with a reachable URL can be tested: production, staging, a Vercel preview, a published Figma Make prototype, a Claude design artifact or a v0 link. Integrations decide when that happens and where the results land.
How Stunt Double plugs in
Three mechanisms connect Stunt Double to your other tools:
- The MCP server at
https://app.stuntdouble.io/api/mcp. Any MCP client (Claude, Claude Code, Cursor, Windsurf, ChatGPT and others) gets the full toolset: create and run checklists, launch persona interviews, triage feedback and comment on pull requests. Client setup lives in the MCP section. - Connections and event triggers. Connect GitHub, Figma, Linear, Slack or Vercel and automations can start on pull requests, deploys, design updates, Linear issues and Slack messages, so checks re-run without anyone asking.
- The dashboard at app.stuntdouble.io, for anything visual: watching session recordings, reading interview reports and managing actors and integrations.
How every connection works
Every integration lives on the Integrations page of your workspace. Open the sidebar, choose Integrations, then click a card. Each one is set up in the same three steps, in the same order, so learning one teaches you all of them.
- 01Your accountConnect the account Stunt Double should use
- 02This workspaceAn owner or admin turns it on
- 03ProjectsEach project picks its file, repository, team or channel
- Your account. Click Connect followed by the tool's name (for example Connect GitHub) and approve Stunt Double on the provider's site. The step then reads Connected as with your account name, and lists the workspaces already using it.
- Your workspace. The second step carries your workspace's name and a switch. An owner or admin turns it on, and the step reads On. Actors, checklists and automations here can use it. Members see Off. Ask a workspace owner or admin to turn it on. until an admin does.
- Projects. Click Choose for a project and pick a project. That opens the project's Settings, on the Integrations tab, where you choose the design file, repository, Linear team, Slack channel or Vercel project it works with. If the workspace has no projects yet, the button reads Create a project.
A step that depends on the one before it stays greyed out until that step is done, so the next thing to do is always the first step without a tick.
Once an integration is on, its page grows a Settings section for anything that applies to the whole workspace, such as GitHub's watched repositories or the Linear agent's identity.
Disconnecting
To remove a connection, open the integration and click Disconnect on the Your account step. A dialog titled Disconnect followed by the tool's name lists every workspace that loses the integration. Confirm, and those workspaces stop using it until someone connects again.
To stop one workspace from using a connection while others keep it, an owner or admin turns off the switch on step two instead. Slack works a little differently: see Slack for Remove versus Disconnect.
Handing setup to someone else
GitHub, Linear and Slack each show an Install link card at the foot of their page. Copy the link and send it to whoever should finish setup: a teammate who owns the GitHub organization, or an admin who has to flip the workspace switch. The link walks them through every step for that workspace and picks up wherever they left off.
Every integration
| Integration | What it does | Who sets it up | What Stunt Double requests |
|---|---|---|---|
| GitHub | Actor pull request reviews, PR comments, automations on GitHub events | Anyone connects; an owner or admin turns it on | OAuth: read:user, user:email, repo. GitHub App: access to the repositories you choose |
| Vercel | Automations on deployment events | Anyone connects, or install from the Vercel Marketplace | The permissions the Vercel integration asks for at install |
| Figma | Design context for actors, automations on file events | Anyone connects; an owner or admin turns it on | current_user:read, file_content:read, file_metadata:read, file_comments:read, file_comments:write |
| Figma plugin | Design reviews, checklists, feedback and actor chat inside Figma | Each designer, from Figma | A plugin token valid for 90 days. The plugin only talks to stuntdouble.io |
| Linear | The Stunt Double agent on Linear issues, automations on issue events | Anyone connects; an owner or admin turns it on and picks the agent's actor | read, write, issues:create, comments:create, app:assignable, app:mentionable |
| Slack | Design reviews in threads, notifications, a workspace agent in direct messages | An owner or admin adds it to the workspace | channels:read, groups:read, channels:history, groups:history, chat:write, chat:write.customize, mcp:connect, commands, users:read, app_mentions:read, incoming-webhook, assistant:write, im:history, im:write |
| Webhooks | Start an automation from any system that can send an HTTP request | Anyone who can edit automations | A secret per automation, sent as X-Webhook-Secret |
| Feedback widget | Collect comments on your live site into a project | Project members; owners and admins set workspace limits | A project key that identifies the project, limited to your allowed domains |
| Self-hosted workers | Run actors on your own network, against sites the public internet cannot reach | Owners and admins, on Enterprise | A worker API key, shown once |
| Desktop app | Use your Mac as a worker | Owners and admins, where workers are available | Your Stunt Double sign-in |
MCP clients (Claude, Claude Code, Cursor, Windsurf, ChatGPT) request
mcp:read, mcp:write and mcp:run. Their setup is in the MCP section.
Guides by workflow
The pages above cover setup. The guides cover what to do once you are connected:
- Design in Claude, validate with personasDesign an artifact, publish it, test it with personas and iterate in the same chat.
- Test Figma prototypes with AI personasWalk prototypes with personas and re-check the built flows when a file changes.
- Idea to evidence with Figma MakePublish a Make site, interview personas against it and feed findings back.
- Verify every change with Claude Code, Cursor and GitHubShip to a preview, verify flows with a checklist and report on the pull request.
- Validate brand, compliance and design systemsCodify a standard as actor knowledge, assert it with checks and re-run on events.
Where each role feels it
- Designers find usability problems while the design is still cheap to change: persona interviews against prototypes, with transcripts and evidence.
- Product managers get research at decision speed: a structured study with a brief, a discussion guide, varied personas and a synthesized report, plus feedback triage that clusters and reproduces what users report.
- Engineers get a user-level regression net: checklists that walk real flows on preview deployments, results as PR comments, and re-runs on every deploy.
- Leadership gets scheduled checks over the critical paths (signup, checkout, the core action), with failures reaching Slack or Linear before they reach support.
- Brand, legal and design-system owners get enforcement instead of periodic review: standards written down once and checked on every deploy or design update.