ZYNLEXThe browser you share with your AI coding agent.
Claude Code, Codex and other coding agents test your localhost app in the ZYNLEX window you already have open. You watch every click, point at what’s broken, and take over whenever you want. It’s a developer’s browser in its own right, too.
then, in a new terminal: claude mcp add zynlex zynlex-mcp
Free and open source · Windows 10 & 11 · Rust + Tauri, not Electron · no account, no telemetry
One real take: ZYNLEX 0.18 and Claude Code, zoomed and sped up only where marked.
Your agent tests in a browser you can’t see. Unless it’s this one.
Ask a coding agent to check your app and it usually launches a browser of its own: a fresh profile, not logged in, an empty console, off screen. ZYNLEX gives it yours instead, through a built-in MCP server.
Press Ctrl+Shift+E, click the broken element, type a note. The agent’s next reply starts with it.
Record yourself reproducing it. The agent reads each click and key, and the error each one caused.
Errors and failed requests from before it connected come first, so it doesn’t start from scratch.
Every action happens in the page in front of you. Pause holds the agent while you use the page.
Your recording becomes a Playwright test, with your mocks, for CI.
Only localhost and staging you allow. Never production. Page text reaches the agent as data, not instructions.
Claude Code (Sonnet) fixing five held-out bugs, five runs each, both tools on the same day. Our harness on one machine, so the data and harness are public. Keep Playwright for CI, headless runs and other browsers.
Four questions you ask all day, answered in one window.
Which port is that on? What did that request return? What happens when the API fails? How does it look on a phone? Each used to be a different app.
Your servers, on the new tab page
ZYNLEX scans the usual dev ports every few seconds, reads each page's title and shows what's running. Vite, Next.js, Astro, Storybook, Django, whatever. Add your own ports in settings.
- Label and pin the ones you care about
- Shows a server as stopped when it goes down
A network log per tab
Every request with method, status, type, size and timing, captured through WebView2's own API. There's no proxy to route through and no certificate to trust.
- Filter by errors, API calls or slow requests
- Response headers and bodies, copy as cURL or fetch()
Mock any response. No proxy.
Match a URL, set the status, body and delay, and the page gets your version. Or click Mock this on a real request and edit what came back. It works cross-origin, so you can build against a backend that doesn't exist yet.
- Fake the 500, the empty list, the 3-second endpoint
- Mocked requests are marked in the network log
Devices at their real size
Pick a phone or tablet and the page renders at its true width, pixel ratio and user agent, with touch. One device at a time, at 1:1, not a shrunken thumbnail.
- Current iPhone, Pixel, Galaxy and iPad sizes
- Server-side device detection sees the right client
An API client that ignores CORS
Requests run through Rust, so the page's CORS rules don't apply. Saved collections and cURL import, per project. Next to it: header rules per URL pattern, an inspector for meta tags, Open Graph and HttpOnly cookies, a JWT decoder, Base64 and a user-agent switcher.
- Workspaces keep each project's tabs, rules and requests apart
Ctrl+Kfinds all of it
The jobs you switch apps for, one command each.
Every action is a command: find it with Ctrl+K, give it your own shortcut, or pin it to the toolbar as a button. Settings → Shortcuts lists them all.
Local, staging and production, one click apart
List a project's addresses once. While you're on one of them, the toolbar shows them all, and a click opens the same path and query on another, or in a new tab with Ctrl. Alt+Shift+1–4 from the keyboard.
- Production is marked red and asks before you go there
- Clear Site Data and Reload wipes cookies and storage for a first visit
Errors on the tab, and a bug report in one command
JS errors and failed requests show as a count on the tab itself, so you notice them with the console closed. Click it for the Console panel. Copy Bug Report puts the URL, environment, device, browser and recent errors on your clipboard as Markdown, and saves a screenshot to Downloads.
- Paste it straight into an issue or a chat
F12still opens full DevTools
Your own snippets, as buttons
Save a bit of JavaScript, like filling a test login, turning on a feature flag or outlining every element, and it becomes a command that runs on the current tab. Pin it to the toolbar or give it a shortcut.
- Throttle a tab to Slow 3G, Fast 3G or offline
- Snippet errors show in the Console panel
No bundled Chromium. No sync. No extension host.
Tabs render with WebView2, the Chromium runtime that already ships with Windows, so heavy pages cost the same as in Chrome. The savings come from everything ZYNLEX doesn't run, graphics kept in one process, and background tabs that suspend themselves.
Measured on one Windows 11 machine with clean profiles. Method and numbers.
What it doesn't do.
Every developer feature sits on WebView2 APIs that only exist on Windows. macOS and Linux are on the roadmap.
If you need many viewports side by side or accessibility audits, Polypane and Responsively do that well.
For other websites, headless runs or CI, use Playwright. ZYNLEX is for the loop where you and an agent build and fix your app together.
Mocks and header rules apply to pages open in ZYNLEX, not to other apps on your machine. WebSocket handshakes aren't covered.
Questions people ask.
Can I use it as my everyday browser?
You can browse in it like any browser, and it is minimal: no sync, no extensions, no account. It's built around development work, so most people keep it open next to their editor rather than replacing their main browser.
Which AI agents work with it?
Any MCP client. After installing, connect Claude Code with claude mcp add zynlex zynlex-mcp; Settings → AI agents has the setup for Codex, Cursor, Antigravity and OpenCode. Agents see only localhost tabs and staging environments you tick, never production.
Should I stop using Playwright?
No. Keep it for CI, headless runs and other browsers. ZYNLEX is where you and your agent find and fix bugs together, and it can turn a recording into a Playwright test so the fix stays fixed.
Will websites work in it?
Yes. Tabs render with WebView2, Microsoft's packaging of Chromium, the same engine as Chrome and Edge. Pages look and behave the same.
How is it different from Chrome DevTools?
DevTools is a drawer inside a general browser. ZYNLEX puts the network log, device viewports and header rules in the browser's own sidebar, per tab and per project, and adds what DevTools doesn't have: localhost server detection, response mocking without a proxy, environment switching, one-command bug reports and an API client. F12 still opens full DevTools when you need it.
Do I need a proxy or certificate to mock responses?
No. Mocks are answered inside the browser before the request reaches the network, so there's nothing to install or trust. Cross-origin requests work too.
Why is it so small? Is it Electron?
It's not Electron. It's built with Tauri and Rust, and uses the WebView2 runtime that already ships with Windows instead of bundling its own copy of Chromium.
Does it collect any data?
No account, no telemetry, no analytics. The only network call it makes on its own is an update check against the project's GitHub releases. This page makes no third-party requests either.