# Stunt Double

> Stunt Double sends AI users through your product to show how real people and AI agents actually experience it, so product, design and growth teams can fix what matters before they ship.

Stunt Double is an AI-powered product testing and research platform. AI actors take on real user personas, browse with their own sessions, email, and knowledge, and surface the insights that only come from genuinely using what you have built. They cover both audiences a modern product serves: the people who use it, and the AI agents that increasingly use it on their behalf.

## When to use Stunt Double

Use Stunt Double when you need evidence of how people, or other AI agents, actually get on with a product in a real browser: what they clicked, where they got stuck, and screenshots to prove it. It works on anything with a reachable URL: production, staging, preview deployments, published Figma or other prototypes.

Best-fit jobs:

- **Check a change before it ships.** A deploy or pull request produced a preview URL and the affected flows need verifying. Create a checklist of observable checks, run it, and read the pass or fail for each with a screenshot and the reasoning (`create_checklist`, `run_checklist`, `get_checklist_run`, then optionally `comment_on_pr`).
- **Find out how users react before building.** A design, prototype or product question depends on user behavior. Run a structured interview with AI persona participants and read the synthesized report (`create_interview`, `add_interview_participant`, `launch_interview`, `get_interview_report`).
- **Keep a critical flow working.** Re-run checklists on a schedule or on deploy, pull request and design events (`create_workflow`, `run_workflow`).
- **Enforce a standard across the product.** Brand and tone of voice, design-system use, accessibility, legal copy, or consistent pricing and terminology across surfaces. Give actors the standard as knowledge and assert it with checks (`add_actor_knowledge`, `create_checklist`).
- **Make sense of stakeholder feedback.** Summarize themes and an implementation brief from comments collected on a project (`list_feedback`, `summarise_feedback`).

Not a fit: code with no URL to visit (use unit or integration tests), load or performance testing, or decisions that need statistically significant evidence from real customers.

## How to call Stunt Double

1. Connect an MCP client to `https://app.stuntdouble.io/api/mcp` (Streamable HTTP). The first connection signs in to a Stunt Double workspace with OAuth 2.1. In Claude Code: `claude mcp add --transport http stunt-double https://app.stuntdouble.io/api/mcp`.
2. Start with `list_workspaces`, then `list_projects`, and reuse existing projects, actors and checklists before creating new ones. Use `search` to find anything in a workspace by name or content.
3. Runs take minutes. Start one, then poll its `get_*_run` or `get_interview_report` tool for the result.

- [llms.txt](https://www.stuntdouble.io/llms.txt): site summary and when to use Stunt Double.
- [Documentation](https://www.stuntdouble.io/support/docs): concepts, tutorials and the API.
- [MCP server docs](https://www.stuntdouble.io/support/docs/api/mcp): connect an agent.
- [MCP manifest](https://www.stuntdouble.io/.well-known/mcp.json): transport, auth and tool list.
- [OpenAPI](https://www.stuntdouble.io/openapi.json): public HTTP endpoints.
- [Sitemap](https://www.stuntdouble.io/sitemap.xml): every public page.

## Getting started

- [Live demo](https://app.stuntdouble.io/demo): free, no sign-up. Runs a real interview against any URL you give it, so it doubles as a sandbox for seeing what a run returns.
- [Sign up](https://app.stuntdouble.io/login): self-serve with email or Google, no sales call. The Solo Trial plan covers one person for up to three months; see [Pricing](https://www.stuntdouble.io/pricing).
- Connecting an MCP client runs the OAuth flow itself (dynamic client registration, PKCE), so there is no API key to request by hand.

## Developer resources

- [Stunt Double API documentation](https://www.stuntdouble.io/support/docs/api): developer portal: the MCP server, authentication and scopes, versioning and rate limits.
- [Stunt Double MCP server reference](https://www.stuntdouble.io/support/docs/api/mcp): connect a client to https://app.stuntdouble.io/api/mcp, with the full tool and prompt list.
- [Stunt Double OpenAPI spec](https://www.stuntdouble.io/openapi.json): OpenAPI 3.1 for the public endpoints and the MCP entry point, with OAuth scopes.
- [auth.md](https://www.stuntdouble.io/auth.md): how an agent registers, gets a token and uses it, in one document.
- [Agent Skills index](https://www.stuntdouble.io/.well-known/agent-skills/index.json): a SKILL.md an agent can install to use Stunt Double well, with its SHA-256 digest.
- [Stunt Double skills and plugin](https://github.com/stunt-double/stuntdouble-mcp): task skills (`npx skills add stunt-double/stuntdouble-mcp`), the Claude Code and Cursor plugins, and the MCP Registry entry `io.stuntdouble/mcp-server`.
- [API catalog](https://www.stuntdouble.io/.well-known/api-catalog): RFC 9727 linkset of every API with its description, docs and health check.
- [Stunt Double API versioning and deprecation policy](https://www.stuntdouble.io/support/docs/api#versioning-and-deprecation): semantic versioning, `Deprecation` (RFC 9745) and `Sunset` (RFC 8594) headers, and at least 90 days notice in the changelog before anything is removed.
- [OAuth 2.1 authorization server metadata](https://app.stuntdouble.io/.well-known/oauth-authorization-server): RFC 8414 metadata: endpoints, PKCE, dynamic client registration and `scopes_supported`.
- [OAuth protected resource metadata](https://app.stuntdouble.io/.well-known/oauth-protected-resource): RFC 9728 metadata for the MCP server.
- [MCP manifest](https://www.stuntdouble.io/.well-known/mcp.json): transports, authentication, scopes and tool list.
- [Public docs MCP server](https://www.stuntdouble.io/.well-known/mcp): no sign-in needed: POST a JSON-RPC `initialize`, then `resources/list` and `resources/read` for these documents.
- [MCP server card](https://www.stuntdouble.io/.well-known/mcp/server-card.json): server-card.schema.json card for the public docs MCP server, with its Streamable HTTP remote.
- [AI Catalog](https://www.stuntdouble.io/.well-known/ai-catalog.json): both MCP servers (the product server and the public docs server), each with the server card on its own host.
- [Public docs agent (A2A)](https://www.stuntdouble.io/.well-known/agent-card.json): no sign-in needed: the Agent Card for an A2A agent at `https://www.stuntdouble.io/api/a2a` that answers questions from these documents.

## Core pages

- [Home](https://www.stuntdouble.io): product overview and what Stunt Double does.
- [Features](https://www.stuntdouble.io/features): every feature, grouped by job, each linked to its documentation.
- [Synthetic research](https://www.stuntdouble.io/synthetic-research): what synthetic research is, how to run a study, and where it does not belong.
- [Pricing](https://www.stuntdouble.io/pricing): plans and pricing.
- [Alternatives](https://www.stuntdouble.io/alternatives): how Stunt Double compares with usability panels, scripted end-to-end tests, manual QA, analytics and synthetic personas, and when it is the wrong tool.
- [Integrations](https://www.stuntdouble.io/integrations): connect Stunt Double to Claude and other tools.
- [Design](https://www.stuntdouble.io/design): how we design Stunt Double, and how product design changes when some users are agents.
- [Guides](https://www.stuntdouble.io/guides): playbooks for wiring Stunt Double into your workflow.
- [Blog](https://www.stuntdouble.io/blog): the Backstage blog.
- [Contact](https://www.stuntdouble.io/contact): get in touch.

## Guides

- [Design in Claude, validate with personas](https://www.stuntdouble.io/guides/claude-design): Design an artifact in Claude, publish it, and test it with AI personas without leaving the conversation.
- [Test Figma prototypes with AI personas](https://www.stuntdouble.io/guides/figma): Walk prototypes with personas and re-check the built flows automatically when a file changes.
- [The fastest idea-to-evidence loop with Figma Make](https://www.stuntdouble.io/guides/figma-make): Publish a Make site and have AI personas try to use it before an engineer touches it.
- [Verify every change with Claude Code, Cursor, and GitHub](https://www.stuntdouble.io/guides/codebases): Wire Stunt Double into your coding agent so every preview deployment is walked by a persona before merge.
- [Collect comments straight from your site](https://www.stuntdouble.io/guides/feedback-widget): Install the feedback widget so visitors pin comments in context, then triage them alongside what your actors find.
- [Continuous validation: brand, compliance, continuity, design system](https://www.stuntdouble.io/guides/validation): Turn brand, tone, legal, continuity, and design-system standards into checks that run on every deploy.

## Use cases

- [Test your signup and onboarding as a new user](https://www.stuntdouble.io/use-cases/onboarding-testing): An AI actor signs up for your product cold, with its own inbox and a real browser, and shows you every step where a new user would stall.
- [Find the step where your checkout loses people](https://www.stuntdouble.io/use-cases/checkout-testing): Actors go through your funnel from landing page to confirmation as the customers you want, and show you the screen where each one stopped.
- [Test apps and prototypes built with AI](https://www.stuntdouble.io/use-cases/ai-built-apps): Publish the app from v0, Figma Make or a Claude artifact and have actors work through it before anyone else does.
- [Check every pull request in a real browser](https://www.stuntdouble.io/use-cases/pull-request-checks): Run a checklist against each preview deployment and post the result on the pull request, so a regression shows up before merge.
- [Test how AI agents get on with your site](https://www.stuntdouble.io/use-cases/agent-readiness): Send an AI agent through your site with a task and see the step where it gives up, before the agents your customers use find it.
- [Check every page against your brand and design system](https://www.stuntdouble.io/use-cases/brand-compliance): Give a project its guidelines and every design review applies them the same way, on live pages and Figma frames, with each finding citing the rule.
- [Test staging and internal tools behind your firewall](https://www.stuntdouble.io/use-cases/internal-tools): Run actors in a browser on your side of the firewall, so staging, admin panels and intranets get tested with no inbound firewall changes.

## Glossary

- [Agent experience (AX)](https://www.stuntdouble.io/glossary/agent-experience): Agent experience is how well an AI agent can use a product on someone’s behalf: find what it needs, understand the page, and finish the task without a person stepping in.
- [AI user testing](https://www.stuntdouble.io/glossary/ai-user-testing): AI user testing sends AI agents with defined personas through a product in place of recruited participants, and reports what each one did and where it got stuck.
- [Browser agent](https://www.stuntdouble.io/glossary/browser-agent): A browser agent is an AI model that operates a web browser: it reads the page, decides on an action such as a click or a keystroke, performs it, and repeats until the task is done.
- [llms.txt](https://www.stuntdouble.io/glossary/llms-txt): llms.txt is a plain-text Markdown file at the root of a website that tells language models what the site is and links to the pages worth reading, in a form that is cheap to fetch and parse.
- [Model Context Protocol (MCP)](https://www.stuntdouble.io/glossary/model-context-protocol): The Model Context Protocol is an open standard for connecting AI assistants to tools and data. A service exposes an MCP server, and any MCP client, such as Claude or Cursor, can call its tools.
- [Prototype testing](https://www.stuntdouble.io/glossary/prototype-testing): Prototype testing puts an unfinished design in front of users to see whether they can complete the key tasks before anything is built.
- [Sycophancy in synthetic research](https://www.stuntdouble.io/glossary/sycophancy): Sycophancy is the tendency of language models to tell the person asking what they seem to want to hear. In synthetic research it shows up as simulated users who like every concept they are shown.
- [Synthetic research](https://www.stuntdouble.io/glossary/synthetic-research): Synthetic research is user research run with AI participants instead of recruited people: interviews, usability tasks and studies where the participants are personas played by AI agents.
- [Synthetic users](https://www.stuntdouble.io/glossary/synthetic-users): Synthetic users are AI agents given a persona (a background, goals and constraints) that stand in for real users in research or testing.
- [Usability testing](https://www.stuntdouble.io/glossary/usability-testing): Usability testing watches people attempt real tasks in a product to find where they hesitate, make mistakes or give up.
- [User persona](https://www.stuntdouble.io/glossary/user-persona): A user persona is a profile of a type of person a product serves: who they are, what they are trying to do, what they know already and what gets in their way.
- [UX regression testing](https://www.stuntdouble.io/glossary/ux-regression-testing): UX regression testing checks, after every change, that the flows users depend on still work and still make sense, not only that the code still passes.
- [Agentic commerce](https://www.stuntdouble.io/glossary/agentic-commerce): Agentic commerce is buying and booking done by an AI agent on a person’s behalf: the agent searches, compares the options and completes the checkout, and the person approves the result.
- [End-to-end testing](https://www.stuntdouble.io/glossary/end-to-end-testing): End-to-end testing runs a scripted path through a whole application, from the interface to the database and back, to check that the parts work together the way a user would meet them.
- [Heuristic evaluation](https://www.stuntdouble.io/glossary/heuristic-evaluation): Heuristic evaluation is an expert review of an interface against a set of principles, such as Nielsen’s ten usability heuristics, that lists each problem found and the principle it breaks.
- [Synthetic monitoring](https://www.stuntdouble.io/glossary/synthetic-monitoring): Synthetic monitoring runs scripted checks against a live site on a schedule, from outside, to catch outages and slow pages before real visitors meet them.
- [Think-aloud protocol](https://www.stuntdouble.io/glossary/think-aloud-protocol): The think-aloud protocol asks a participant to say what they are thinking as they work through a task, so a researcher hears the reasoning behind each click, not only the click.

## Legal

- [Privacy policy](https://www.stuntdouble.io/privacy): how we handle personal data.
- [Terms of service](https://www.stuntdouble.io/terms): terms governing use of Stunt Double.
- [Security](https://www.stuntdouble.io/security): our security posture and how to report a vulnerability.

## Recent blog posts

- [Introducing the Stunt Double Index](https://www.stuntdouble.io/blog/introducing-the-stunt-double-index): A free, public benchmark of how AI agents experience websites. Look up your domain, see where agents stop, and get the list of what to fix.
- [Your Users may not be Human](https://www.stuntdouble.io/blog/your-users-may-not-be-human): In June 2026 Cloudflare confirmed machines now generate more web traffic than people. Every one of those visits is a user nobody has watched, and nobody can ask why. What does user-centred mean in this new age of AI?
- [Building with Supabase](https://www.stuntdouble.io/blog/building-stunt-double-on-supabase): How we run a multi-tenant AI agent platform on Supabase: row level security for tenant isolation, OAuth and passkeys for humans and machines, pgvector for agent RAG, read replicas for global reads, and preview branches for every pull request.
- [What Is MCP and Why It Matters for Product Teams](https://www.stuntdouble.io/blog/what-is-mcp-and-why-it-matters-for-product-teams): The Model Context Protocol lets AI assistants call tools in external services. For product teams, that means testing, feedback, and issue tracking without leaving your editor.
- [Building a Self-Healing Product Feedback Loop](https://www.stuntdouble.io/blog/self-healing-feedback-loop): Turn error signals and feedback into almost instant product improvements with triage and cloud coding agents.
- [For Humans & Their Machines](https://www.stuntdouble.io/blog/for-humans-and-their-machines): User-centred design in a world with AI agents.

## Optional

- [Stunt Double Index](https://index.stuntdouble.io): independent benchmark of how AI agents experience the web.
- [Status](https://status.stuntdouble.io): service status.