MyTest

Still looking for QA to test?

Your QA team is one prompt away.

MyTest turns Claude Code, Claude Desktop and Antigravity into a tracked, versioned QA team. Create agents by talking. Watch every click. Ship only verified bugs.

Start testingSee how it works

One URL to connect. No SDK. No scripts to maintain.

  • Claude Code
  • Claude Desktop
  • Antigravity
  • PagePilot
  • Playwright MCP
  • Model Context Protocol

The problem with QA today

Still writing test scripts nobody maintains? Still waiting for a QA slot before every release? Still chasing flaky failures that were never real bugs? Your agents already know how to click through your app. They just have nowhere to report.

MyTest is where they report.

How it works

From “hello” to a verified bug list in four steps.

  1. 01

    Connect

    Paste one URL into Claude Code, Claude Desktop or Antigravity. Approve once in your browser. No SDK, no keys to copy.

  2. 02

    Set up

    Run /mytest-onboard and your agent interviews you: what the app is, which flows matter, what must never happen. Every project starts with 21 ready agents, a browser agent for PagePilot included.

  3. 03

    Run

    Say “run a smoke test on staging”. One call starts the run with an operator and an independent verifier. Saved scenarios make it repeatable.

  4. 04

    Trust

    Every failure carries a screenshot, console and network facts, and a second agent's independent reproduction. The verdict is checked against the evidence, not taken on faith.

A real session

You talk. The agents test. MyTest keeps the receipts.

One sentence creates the project, the environment, the agent and the run. Two agents split the work, one finds a problem, and it only counts once it reproduces.

Try it yourself
claude · illustrative session
you ›/mytest-test Check that a new design appears in the designs table after the form is submitted.
mytest.get_project_context✓ brief, 6 scenarios, 3 known issues
mytest.run_test✓ RUN-12 · browser-agent + qa-verifier
mytest.get_run_credentials✓ staging test account (never shown in chat)
browser-agentopened /designs/new, filled 6 fields, submitted
browser-agentnew row missing from the table · screenshot attached
mytest.report_bug✓ BUG-004 candidate · HIGH · 3 steps · evidence
qa-verifierreproduced from a clean session, twice
mytest.complete_run✓ FAILED · 1 verified bug · 5 of 6 scenarios passed

Run architecture

Design the team for every run.

Need a form filler and a separate checker? Ask for it. Each role gets only the tools it needs, handoffs are explicit, and saved agents can be reused as roles. The run brief tells your AI tool exactly how to execute it.

  1. 01 · role

    mytest-manager

    Plans the run, delegates, decides an honest verdict

    tools: MyTest + subagents

  2. 02 · role

    browser-agent

    Executes the steps and captures evidence

    tools: PagePilot: snapshot · navigate · click · fill_form · screenshot

  3. 03 · role

    qa-verifier

    Independently reproduces every bug before it counts

    tools: PagePilot (fresh session) + verify_bug

Features

Everything a QA lead wants. Nothing to install.

Talk to it

55 MCP tools and slash commands. Set up projects, save scenarios, run tests, inspect results and export bugs from the chat you already have open.

A QA team on day one

21 default agents in every project: a manager, a PagePilot browser operator, an independent verifier, and specialists for login, forms, smoke checks, regressions and exploration.

Verified means verified

MyTest is the referee. Bug reports need steps, expected and actual results. Duplicates and known false positives are caught. A bug counts once a different agent reproduces it.

Scenarios and coverage

Save repeatable test cases with priorities. Every run records PASS, FAIL, BLOCKED or SKIPPED per scenario, so you get coverage and a pass rate that means something.

Evidence, not claims

Screenshots and console or network logs are attached to failures and shown right beside the bug. You can check what the agent saw.

Memory that compounds

A Project Brief you write, and facts agents propose that you approve. Every run starts knowing your app, your quirks and what is not a bug.

Honest verdicts

PASSED, FAILED, BLOCKED or INCONCLUSIVE. A missing test account is BLOCKED, not a product failure, and a PASS over a failing scenario is refused.

Built for teams

Invite testers and developers, switch organizations, get notified in Slack, and export bugs as CSV or a ready-to-paste issue. Test accounts live in an encrypted vault.

Built for people who ship.

Solo founders

Ship on Friday without a QA hire. Describe what matters, let agents walk your app, and get a short list of bugs that are real.

Product teams

Give every squad a reusable test crew. Agents live inside the project, versions are tracked, and runs stay comparable from release to release.

QA leads

Direct testers instead of writing scripts. Design the team for each run, restrict the tools every role can touch, and audit every action.

Just say it

Your whole QA workflow, in plain English.

  • “/mytest-onboard”
  • “Create a project called Checkout with a staging environment at https://staging.acme.dev.”
  • “Design scenarios for our sign-up form, including invalid emails and empty fields, and save them.”
  • “/mytest-test Check that a new design appears in the table after the form is submitted.”
  • “Run all saved scenarios on staging and tell me what failed.”
  • “What happened in RUN-12? Show me the verified bugs and the screenshots.”
  • “Remember that the date picker uses US format, so we stop reporting it.”
MCP tools
55
default agents
21
verdicts, server-checked
4
URL to connect
1

Stop maintaining tests. Start directing testers.

Before

  • Write and maintain brittle scripts
  • Wait for QA capacity
  • Triage flaky noise and duplicates
  • Screenshots pasted into chat

With MyTest

  • Describe the test in a sentence
  • Agents run the moment you ask
  • Bugs an independent agent reproduced
  • Evidence, coverage and a verdict you can trust

Security

Trust is a feature, not a footnote.

OAuth 2.1 with PKCE

Connect through your browser. Access tokens are short-lived and refresh tokens rotate, with reuse detection that revokes the grant.

Test accounts in an encrypted vault

Passwords are stored with AES-256-GCM and released only to a running run for its own environment. Credentials typed into free text are redacted before they are stored.

Roles, isolation and two-factor

Every query is scoped to your organization. Owners, admins, members and viewers see and change different things. Sign-in supports authenticator-app two-factor.

Every action on the record

Sign-ins, agent calls, credential releases, exports and refusals are audited with time, actor and IP. Admins see the whole organization.

Every action, on the record.

Sign-ins, agent calls, connections, exports and refusals land in an audit feed with time, actor and IP. Admins see the whole organization; everyone else sees their own activity.

  • Dana connected an agent hostOAuth · 2m agoSuccess
  • Claude Code called create_runAgent · 2m agoSuccess
  • Unknown was refused at the MCP endpointAgent · 1h agoDenied

Questions, answered.

What is MyTest?

MyTest is a control plane for AI QA agents. You connect it to Claude Code, Claude Desktop or Antigravity over the Model Context Protocol, then set up projects, run tests and review verified bugs with evidence, all from chat or from the dashboard.

Does MyTest run my tests in the cloud?

No. Agents run locally inside the AI tool you already use. MyTest stores the agent definitions, scenarios and memory, coordinates runs and records what happened, so it never sits between your agent and your browser.

Does it work with PagePilot?

Yes. The default browser agent is built for PagePilot and uses its snapshot, navigate, click, form and evidence tools. You can switch a project to Playwright MCP or another browser MCP in the project settings, and the same agents adapt.

How does MyTest keep agents honest?

The server checks the work. Bug reports need reproducible steps, expected and actual results, and a URL. Duplicates and known false positives are recognised. A bug counts once a different agent reproduces it. A run cannot be declared PASSED over a failing scenario or a verified bug, and tool use outside a role's allowlist is detected and recorded.

What is project memory?

A Project Brief written by your team, plus facts agents learn, such as quirks, known issues and test data. Agents only propose new facts; a person approves them. Rejected bugs are remembered as not-a-bug, so they are not reported again.

Are test account passwords safe?

Test accounts are entered by a person in the web app and stored encrypted at rest. An agent can read one only while a run is active, for that run's environment, and every release is audited. Passwords pasted into objectives or notes are automatically redacted.

Which AI tools does it work with?

Claude Code, Claude Desktop and Antigravity, plus any MCP client that supports remote HTTP servers. Connecting takes one URL and one approval in your browser.

Can my whole team use it?

Yes. Invite testers and developers to your organization with an invitation link tied to their email, assign roles from owner to viewer, and switch between organizations. The audit log shows who did what.

What counts as a verified bug?

An agent first reports a suspected issue as a candidate with steps and evidence. A different role then reproduces it from a clean state and verifies or rejects it. Severe bugs verified only by their reporter wait for a person to confirm. Only verified bugs reach your default list and CSV export.

Stop looking for QA.

Start a run. Your first verified bug is one conversation away.

Create your accountSign in