Keywright is a desktop IDE for Robot Framework web tests using Browser library. Record a browser interaction as Robot keywords, edit the test, run it, and inspect the failed keyword, screenshot or Playwright trace from the same project.
Setup · First run · User guide · Troubleshooting · Development
| Task | What Keywright provides |
|---|---|
| Write a browser test | Test Generator records actions and assertions as Browser keywords. Pick a locator or record steps directly into an open file. |
| Maintain a suite | Monaco editor with keyword completion and documentation, navigation, references, rename, quick fixes, argument checks and Robocop integration. |
| Review and commit changes | Source Control shows Git diffs and lets you stage files and commit. |
| Run a test or suite | Test selection, tag filters, Robot arguments, dry run, streamed keyword results and re-runs when watched files change. |
| Investigate a failure | Source navigation, Browser screenshots, Robot logs and reports, context traces, breakpoints, stepping and variable inspection. |
| Compare results | A local history of the last 30 run summaries, including newly failing, fixed and slower tests. |
| Work with an AI client | An optional MCP server for Claude Code, Claude Desktop, Codex or Cursor. The client edits project files, runs tests, and explores the application in its own browser, where each action yields the matching Browser step. |
Tests remain ordinary .robot and .resource files. They can be edited in another
editor and run with the project's Python outside Keywright. Recorded code is a
starting point: review selectors and assertions, and move repeated behavior into
user keywords as the suite grows.
Keywright is in early development, currently version 0.6.0. Run it from source while the project is being tested and refined. Installer releases are on hold until they can be code-signed. Settings and behavior may change before 1.0.
- Platform: Windows 10/11. Development and integration tests currently target Windows; other desktop platforms have not been verified.
- Browser integration: Browser library. The recorder, locator picker, screenshots and trace integration target its Playwright backend.
- SeleniumLibrary: not a supported integration. Keywright does not generate SeleniumLibrary tests or translate existing SeleniumLibrary suites.
- Projects: a project-local
.venvby default, or an explicit Python path inkeywright.json. Existing layouts can configure the test and results directories.
How dependable each part is today:
| Area | Status |
|---|---|
| Opening projects, editing, running tests, results with screenshots, save conflicts and recovery of unsaved files | Exercised by desktop UI tests in CI on every push |
| Test Generator and locator picker | Built on Playwright's private recorder API. Playwright is pinned to a tested version because an update can change that API |
| Installer and automatic updates | Not available. They return once releases can be code-signed |
| macOS and Linux | Not tested |
See the guide for configuration and current limitations, including saved-file execution and trace availability.
Use Windows 10/11. The setup checks these tools before installing app dependencies:
| Tool | Version to use | Check in PowerShell |
|---|---|---|
| Node.js with npm | 24.x or 26.x LTS | node --version and npm.cmd --version |
| Python | 3.13.x, available through the Windows py launcher |
py -3.13 --version |
| Git | Required for Source Control and for cloning or updating a checkout | git --version |
If Node.js or Python is missing or unsupported, setup offers to install a supported
version through WinGet and asks for confirmation with y/n. If WinGet is
unavailable, it shows the official download link. Exact tested package versions are listed in
Contributing.
Open PowerShell in the folder where you want to clone Keywright, then run:
git clone https://github.com/silvestrodecaro/keywright.git
cd keywright
.\setup.batWithout Git, use Code → Download ZIP on GitHub, extract the archive and open
setup.bat in the extracted folder. The setup works from another terminal folder
too: it finds its own repository directory.
Setup installs the locked app dependencies, Electron and Chromium, prepares the
included example's Python environment, builds the editor and runs its two browser
tests. It also reports dependency audit findings. The full output is saved under
out/setup/; a failed step stops setup and shows the error and log path.
See Troubleshooting for installation and launch errors.
When setup finishes, choose whether to open Keywright. To try the prepared example,
choose Open folder…, select examples/browser-walkthrough inside the clone,
then select Testing → Run all. Both tests should pass.
Browser executables are stored in the user cache, outside project virtual environments. Projects can stay in their chosen folders, including nested paths and paths containing spaces. Each Robot project still has its own Python environment.
For later launches, open PowerShell in the same keywright folder and run
npm.cmd start. Setup and start show progress and timings; launch logs are in
out/start/. Expand Update Keywright on the welcome screen for
update instructions.
See manual setup for the individual commands.
To create your own test project after trying the included example:
- Choose New project, enter a name such as
my-web-testsand select a writable parent folder. - Keep virtual environment and browser installation enabled, then choose
Create project. The wizard creates
.venv, installs the Python packages and downloads Browser's browsers. Wait for the setup steps to finish, then choose Open project. If a step fails, its output remains visible and the folder is kept: see repair a partial setup. - Select Testing → Run all. The generated project contains one test, so the
expected result is one passing test. It opens
https://example.comand requires network access. The footer should show the Python executable inside this test project's.venv.
To try recording and failure analysis with a supplied local page, follow Record a test and investigate a failure. That example has two tests and works offline after its dependencies and browsers are installed.
Already have Robot tests? Choose Open folder… and select their project folder. If they need an environment or use a different layout, follow open an existing project. Opening a folder reads its files; environment setup is a separate step.
| Document | Contents |
|---|---|
| Walkthrough | Run the supplied example, record a test, reuse a resource and inspect a failure |
| User guide | Projects, editor, recording, execution, debugging, configuration and shortcuts |
| Troubleshooting | Environment setup, discovery, browser launch, missing traces and editor issues |
| Contributing | Development setup, tests, integration prerequisites and dependency review |
| Architecture | Electron/Python boundaries, application flows, persistence and integration contracts |
| Changelog | Changes in each version |
Project files and Robot output are stored in the project directories. Run summaries are kept in the configured results folder; recent projects, interface settings and copies of unsaved files are stored in Electron's user-data directory. Keywright has no analytics or telemetry integration in the application source.
The MCP server is off by default. When you turn it on, Keywright accepts
connections on 127.0.0.1 only, with a token. A connected AI client sends what it
reads, including project files, page snapshots and test results, to its own service
under that client's terms.
Source runs do not check for application updates. Project setup downloads Python packages and Browser dependencies; browser installation downloads browser binaries. Recorded pages and executed tests make the network requests their applications and test code require. Logs, screenshots and traces may contain data from the system under test; review them before sharing.
Report a reproducible problem or propose a change in the issue tracker. Include the Keywright version, project Python and relevant package versions, the steps to reproduce, and a small example with sensitive data removed. Report security vulnerabilities privately as described in SECURITY.md. Ask questions and share ideas in Discussions. Participation is covered by the Code of Conduct.
Development instructions are in CONTRIBUTING.md. Keywright is available under the MIT license.
