TestDriver
Cover hard-to-test application flows with vision-driven end-to-end QA tests.
- Source repo
- testdriverai/testdriverai
- Stars
- ★ 344
- Last updated
- 2mo ago
- Primary language
- JavaScript
- FA score
- 55/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specific
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A QA team validating desktop file drag-and-drop can drag a document icon to the Recycle Bin and assert that it is no longer visible.
- Not a fit if
- Teams that do not want tests running in TestDriver cloud sandboxes
- Teams that do not use JavaScript or Vitest
- Website
- testdriver.ai ↗ set on GitHub
- Maintenance
- Commits in the last 6 monthslast commit 2026-09-21latest release v7.4.5 (2026-02-27)
- Source review
- 55/100 · Major gaps
What does this agent do, and when should you use it?
TestDriver is a JavaScript plugin for Vitest that supports end-to-end QA tests. Tests use testdriver.find() to locate screen content, perform mouse actions, and use testdriver.assert() to check visible state. Its quick start launches a cloud sandbox, opens Chrome, and runs tests using AI vision. Results include a video replay. The README lists third-party web apps, desktop apps, browser and VS Code extensions, and flows involving iframes, canvas, video, and file uploads among its use cases.
Run npx testdriverai init in a project to initialize dependencies, the API key, MCP configuration, and an example test; then execute tests with Vitest. Test code can call testdriver.find("...") to locate visible text or icons and use mouse methods such as mouseDown() and mouseUp(). testdriver.assert(...) checks a described visible state. The documented quick-start flow creates a cloud sandbox, launches Chrome, runs the vision-based test, and returns a video replay.
- A QA team validating desktop file drag-and-drop can drag a document icon to the Recycle Bin and assert that it is no longer visible.
- Engineers maintaining third-party web app integrations can cover end-to-end flows that are difficult with conventional browser automation.
- Teams testing VS Code or Chrome extensions can use visual interaction to verify actions in extension interfaces.
- Teams checking OAuth flows or AI chatbots can run interactions and assert results based on visible screen content.
- Teams validating PDFs, file uploads, spelling and grammar, or OS accessibility can include those flows in end-to-end tests.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npx testdriverai init
How do you install or deploy this agent?
A TestDriver account and API key are required. The README's initialization command installs Vitest and TestDriver, sets the API key, configures MCP for an AI assistant, and generates an example test.
npx testdriverai initHow do you use this agent?
Write a Vitest test in the project, use testdriver.find() to locate screen content and interact with it, then verify the result with testdriver.assert(). Run tests with:
vitest runInteraction and assertion from the README example:
await testdriver.find("New text document").mouseDown();
await testdriver.find("Recycle Bin icon").mouseUp();
const result = await testdriver.assert(
'the "New Text Document" icon is not visible on the Desktop'
);
expect(result).toBeTruthy();What are this agent's strengths and limitations?
- The Vitest plugin form puts vision-based interaction tests in a JavaScript test project.
- Screen-content targeting and mouse actions support interactions such as the documented desktop icon drag-and-drop.
- The quick start documents cloud sandbox execution, Chrome, and video replay for reviewing test runs.
- Listed use cases include desktop apps, extensions, OAuth, PDFs, file systems, and canvas workflows.
- Execution depends on a TestDriver account, API key, and cloud sandbox; the README does not establish an offline mode.
- The README describes a JavaScript plugin for Vitest and does not document direct support for other test runners.
- The supplied material does not give pricing details, data-retention terms, or cloud execution region options.
- Stability metrics, retry behavior, and failure-diagnostics details for visual targeting and assertions are not described.
How does this agent compare with similar options?
The README positions TestDriver for end-to-end flows that are hard to cover with other tools and names Cypress, Playwright, and Selenium as related testing tools; it does not provide feature or cost comparisons.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| TestDriver This agent | 55 · Major gaps | Hosted serviceFreemium | ★ 344 | 2mo ago | JavaScript | — |
| Alysis Code | 72 · Some gaps | CLIFreemium | ★ 229 | today | Python | OpenAI API · Claude API |
| Interceptor | 69 · Some gaps | CLIFree | ★ 638 | 1d ago | TypeScript | Codex · Claude Code · Claude.ai |
| Kane CLI | 67 · Some gaps | CLIFreemium | ★ 267 | 1d ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The materials show API-key sign-in, a telemetry opt-out, and GitHub PR comments that do not block merges. However, the MCP server enables Sentry by default with `sendDefaultPii: true`, and the packet does not explain collected fields or retention, so least privilege, data flow, and sensitive-data handling are only partially evidenced. Cloud sandboxes, browser actions, AWS deployment, and automated PR creation have external effects; prior confirmation and recovery are not adequately described. Dependencies have version ranges but no audit or vulnerability-mitigation evidence. Repository and product attribution are reasonably clear; publisher identity remains unknown.
The SDK, MCP, Vitest, and older CLI documentation conflict on versions and support: for example, the installation draft claims Jest/Mocha support while the manual guide says only Vitest; the CLI quickstart requires Node 20.19 while another installation draft says Node 18. This lowers consistency. The packet does not establish dependency or cloud-service availability. It provides troubleshooting for API keys, missing modules, timeouts, and network failures, but does not cover a broader range of failure cases.
The materials clearly cover web and desktop apps, extensions, OAuth, uploads, and other test scenarios, and explain when to use AI exploration versus explicit actions. The GitHub Agent is limited to writing and maintaining tests and cannot change application code; however, the SDK's `ai()` accepts broad natural-language multi-step tasks, so capability boundaries still require care. Triggers are distinguished across PRs, issues, and @mentions. The SDK depends on cloud services, Vitest, Node versions, and optional AWS setup, with conflicting environment guidance in the packet.
The README, skills, CLI quickstart, manual guide, and examples provide several entry points, but the older installation draft conflicts with current guidance. Installation, configuration, API-key handling, and common troubleshooting examples are fairly complete. TestDriver naming is stable, though skills and multiple interfaces coexist. Some limitations are documented, but not comprehensively. Package metadata specifies ISC; a version is present, but no changelog is included. Maintenance responsibility is only partly clear: the repository, generated-doc markers, and product site are identified, but the author field is empty and no maintenance commitment or support period is stated.
Natural-language actions, visual assertions, replays, GitHub test automation, and concrete code examples make the output useful for authoring and debugging end-to-end tests. Visual targeting and AI exploration offer clear value over hand-written scripts, but the materials also say AI can be slower, and hosted sandboxes, device minutes, and external APIs incur costs that are not sufficiently quantified.
Most feature claims have corresponding explanations in skills, configuration, or examples. Conflicting SDK-support guidance between the README and docs weakens cross-source corroboration. The Agent skill distinguishes observed findings from suspected ones and says an unsigned installation can only review statically; this packet itself is a static sample and cannot validate runtime outcomes. The materials make a reasonable distinction between claims and execution evidence.
- The MCP server enables Sentry by default and sets `sendDefaultPii: true`; confirm telemetry contents and organizational policy before using credentials or real application data.
- An AWS example uses `AllowedIngressCidr=0.0.0.0/0`, although the docs warn to restrict source IPs; do not copy that example into production unchanged.
- Automated tests may start cloud sandboxes, operate an application, record and upload replays, and create PRs containing test code; begin with isolated test accounts and non-sensitive data.
- Node and Vitest support guidance conflicts across the packet; verify the requirements for the target release before installation.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's Changed * Ian/s3 by @ianjennings in https://github.com/testdriverai/testdriverai/pull/701 Full Changelog: https://github.com/testdriverai/testdriverai/compare/v7.3.43...v7.4.5View on GitHub ↗ -
Release notes excerpt
Full Changelog: https://github.com/testdriverai/testdriverai/compare/v7.3.43...v7.3.44View on GitHub ↗ -
Release notes excerpt
Full Changelog: https://github.com/testdriverai/testdriverai/compare/v7.3.42...v7.3.43View on GitHub ↗
FAQ
What do I need to get started?
npx testdriverai init.Where do the tests run?
What kinds of flows can it test?
Can I use a test runner other than Vitest?
How much does it cost?
Is TestDriver open source?
Is TestDriver safe to use?
Is TestDriver still maintained?
Who makes TestDriver?
What is TestDriver's official website?
Thanks — we use this to improve the page.