MCP for QA: How to Trigger User Journey Tests from Claude

The Problem: Testing Lives in a Separate Dashboard

I hate context switching. You're working in Claude or Cursor, building a feature, and you want to test whether your new sign-up flow actually works. So you tab over to your QA dashboard, find the right workspace, click the checklist, wait for results, then flip back to your editor. It's five minutes of friction every time.

What if you didn't have to leave Claude?

That's where MCP comes in. The Model Context Protocol lets you trigger user journey tests directly from your editor or terminal, without ever opening a separate application. Just say "run the sign-up checklist" and Claude will do it, then report back with results and findings.

What MCP Does for Testing

If you're new to MCP, it's a protocol that lets AI assistants (Claude, Cursor, etc.) call tools in external services. For QA, that means your language model can:

• List workspaces and actors
• Run checklists against your live product
• Fetch test results with screenshots and friction points
• Create issues in Linear based on findings
• Integrate with your existing development workflow

No API keys to manage in your code, no separate dashboards to monitor. The test infrastructure just becomes another tool available in your editor.

Setting Up Stunt Double's MCP Server

Connecting to Stunt Double's MCP server takes about 60 seconds.

Claude Code (via terminal)

claude mcp add --transport http stuntdouble https://app.stuntdouble.io/api/mcp

Claude Code will open a browser window asking you to sign in with your Stunt Double account. Click through and you're done.

Claude Desktop

1. Open Claude Desktop
2. Click Settings → Connectors → Add
3. Paste: https://app.stuntdouble.io/api/mcp
4. Sign in via OAuth when the browser opens

Cursor / Windsurf

Add the following to your .cursor/mcp.json (or create it if it doesn't exist):

{
  "mcpServers": {
    "stuntdouble": {
      "url": "https://app.stuntdouble.io/api/mcp"
    }
  }
}

Restart Cursor, and the tools will be available in your composer and chat.

Auth: The first time you use a tool, your browser opens automatically for OAuth 2.1 authentication. Sign in with your Stunt Double account, then return to your editor – the session persists across tool calls.

Example: Testing Sign-Up from Claude

Here's what a real QA workflow looks like. Let me walk you through it: you're in Claude and want to verify the sign-up flow works.

Step 1: List Your Workspaces

You: "List my workspaces"

Claude:
- Marketing (workspace-slug: stunt-double-marketing)
- Product (workspace-slug: stunt-double-product)

Step 2: Get Your Actors

You: "Show me the actors in the Product workspace"

Claude:
- First-time user (persona: no prior account, UK-based)
- Power user (persona: experienced SaaS user, optimising workflow)
- Accessibility-focused (persona: screen reader user, keyboard nav)

Step 3: Run the Sign-Up Checklist

You: "Run the sign-up checklist for the Product workspace"

Claude:
Checklist run initiated: run_id = chk_run_abc123
Status: In progress...
[After a few seconds]
Status: Complete

Results:
✅ Create account (passed)
✅ Verify email (passed)
⚠️ Navigate to dashboard (warning: 2.4 seconds, expected <2s)
✅ View workspace settings (passed)

Step 4: Get Detailed Results

You: "What did the sign-up checklist find?"

Claude:
Checklist: Sign-up Flow
Run: chk_run_abc123
Status: Mostly passing

Findings:
1. Email verification is working, but verification email arrived after 1.8 seconds
2. Dashboard navigation is slightly slow, took 2.4 seconds instead of 2s target
3. First-time user found the workspace settings button quickly
4. No friction points detected in form validation

Recommendations:
- Consider optimising dashboard load time (query caching?)
- Email delivery is acceptable but could be faster
- Overall UX is smooth

Real-World Example: Catching a Bug Before Deployment

Here's how this plays out in practice. You're about to ship a checkout update and you ask Claude:

"Run the purchase flow checklist before we deploy"

Claude triggers the checklist. A minute later:

❌ Payment processing (failed)
  Error: "Card field not found. Unable to locate payment input."

Payment is processed through Stripe Elements (iframe).
The card field isn't labelled correctly for automated testing.
Agent couldn't interact with it.

Boom. You've found the issue before it hits production. You fix the iframe labelling, ask Claude to run it again, and confirm it passes. All without leaving your editor.

Integrating Results into Your Workflow

Test results don't live in isolation; they flow into your development tools:

Create Issues in Linear

After running a checklist, you can ask Claude:

"Create a Linear issue from these findings with the high-priority label"

Claude will create an issue in your Linear workspace with the test findings, screenshots, and reproduction details.

Track in GitHub

Checklist results can be posted as PR comments or issue descriptions, giving your team visibility into test status during code review.

Triage Feedback

If your checklist surfaced user feedback or friction points, Stunt Double's feedback management tools let you triage directly from Claude.

Available MCP Tools

The Stunt Double MCP server exposes these tools:

• list_workspaces –Get all your workspaces
• get_workspace –Get workspace details
• list_actors –List actors in a workspace
• get_actor –Get actor details and persona
• create_actor –Create a new AI actor
• list_checklists –List all checklists
• run_checklist –Trigger a checklist run
• get_checklist_run –Fetch results of a completed run
• list_workflows –List multi-step workflows
• run_workflow –Trigger a workflow
• list_feedback –Get feedback submissions
• get_feedback –Get feedback details

See the full tool docs at https://stuntdouble.io/docs/mcp for detailed parameter and response schemas.

What's Next: CI/CD Integration

Once you're comfortable running tests from Claude, the next step is automating them in CI/CD:

• Pre-deploy checks: Run critical checklists on every deploy to staging. If the sign-up flow breaks, you know before it hits production.
• PR checks: Run checklists as part of your GitHub Actions workflow. Results appear as PR comments.
• Nightly regression: Set up scheduled checklist runs. If something breaks overnight, you'll have the report in Linear by morning.

This is where AI-driven QA scales. You're not just testing manually when you remember –you're testing continuously, with results flowing into your normal development process.

The Real Win: No Context Switching

What I love about this setup is that testing becomes part of your natural development flow. You don't need a separate QA dashboard, separate credentials, separate context – you just ask Claude and it happens.

Trigger tests from your editor. Check results. Create issues. Get back to building. That's the real beauty of MCP for QA.

Getting Started

Ready to give it a go? Here's all you need to do:

1. Add the MCP server to Claude Code, Desktop, or Cursor
2. Sign in via OAuth
3. Ask Claude to "list my workspaces"
4. Run a checklist: "Test the sign-up flow"
5. Review results and act on findings

The full setup takes less than a minute, and once you've run your first test from Claude, you'll wonder how you ever worked without it.