How to give Claude Code or Codex a browser to test your app in 2026: 9 ways compared

Disclosure: Enjoy is our product, and its agents test in a browser of its own. We checked every way below against its maker’s own pages on November 8, 2026, and we say where another way is the better pick, including the free ones that don’t need Enjoy.

The ways to give an agent a browser, compared

An agent that can’t see the page guesses: it writes the CSS, says the button works, and leaves the checking to you. Give it a browser and it can open the page, click through the flow, read the console and fix what it finds before it calls the work done. Anthropic’s Chrome guide puts the loop plainly: “read console errors and DOM state directly, then fix the code that caused them.” The ways below differ in whose browser the agent drives, yours with your sign-ins or a clean one of its own, and in which agents can use it.

Way	Best for	Price, checked November 8, 2026	Browser it drives	Agents
Claude in Chrome	Claude Code on sites you’re signed in to	Included in Claude Pro ($20 a month), Max ($100), Ultra ($200), Team and Enterprise	Your Chrome, Edge, or another Chromium browser	Claude Code
The Claude desktop app’s Browser pane	Claude Code checking its own changes, with no setup	Included in every paid Claude plan	A clean built-in browser; Mac and Windows, Linux in beta	Claude Code
The ChatGPT app’s browser	Codex, with comments you leave on the page	Listed from Plus ($20 a month) in Codex’s pricing	A built-in browser, or your own through its extension	Codex
Playwright MCP	Any MCP agent, in Chrome, Firefox, WebKit or Edge	Free, Apache 2.0	Its own browser, headed by default, or your tabs	Claude Code, Codex and any MCP client
Playwright CLI	Coding agents that need their context for the code	Free, Apache 2.0	Its own browser, headless by default	Claude Code, Copilot and any agent that runs commands
Chrome DevTools MCP	Debugging: console, network and performance	Free, Apache 2.0	Chrome	Claude Code, Codex, Antigravity, Cursor, Copilot and other MCP clients
agent-browser	A fast command-line browser for any agent	Free, Apache 2.0	Chrome for Testing, or your running Chrome	Any agent that runs commands
Orca	A browser beside each worktree, free	Free, MIT	A Chromium pane in the app; Mac, Windows and Linux	Claude Code, Codex and any agent CLI
Enjoy	Every lab’s agent testing beside the conversation, with nothing to set up	A monthly subscription, free for 7 days (pricing)	A built-in browser; Mac, Windows and Linux, screenshots on Mac	Claude Code, Codex, Grok Build, Antigravity, Muse Code, OpenCode

Prices are per month as each maker lists them, before tax. The free tools run on your own Claude, ChatGPT or other plan, which its maker bills for separately. Claude’s pricing has Pro at $20 a month, or $17 billed yearly, and Codex’s puts Codex in every ChatGPT plan, with Plus at $20 and Pro from $100.

Claude in Chrome

Best for: Claude Code users who want the agent to check sites they’re already signed in to.

Start Claude Code with claude --chrome, and it drives your own browser through the Claude in Chrome extension: “Claude opens new tabs for browser tasks and shares your browser’s login state, so it can access any site you’re already signed into. Browser actions run in a visible Chrome window in real time.” It works with Chrome and Edge, and Claude Code “also detects the extension and sets up the connection in other Chromium-based browsers, including Brave, Arc, Vivaldi, and Opera.” Anthropic lists what it’s for: “test form validation, check for visual regressions, or verify user flows.” When it reaches a login page or a CAPTCHA, “it pauses and asks you to handle it manually.” Run /chrome to check the connection or turn it on for every session; in the VS Code extension, type @browser.

Price: included in Pro, Max, Ultra, Team and Enterprise. Claude’s pricing lists Claude in Chrome on Pro, Max and Ultra, but not on Free.

Watch for: it needs “A direct Anthropic plan,” and “If you authenticate with an API key or a long-lived token from claude setup-token, Claude Code keeps Chrome integration off, even when you pass --chrome.” It “isn’t supported in Windows Subsystem for Linux (WSL).” Turning it on for every session “increases context usage since the browser tools and their instructions are always loaded,” and in long sessions “The Chrome extension’s service worker can go idle,” which breaks the connection until you reconnect it from /chrome.

The Claude desktop app’s Browser pane

Best for: people who run Claude Code in Anthropic’s own app and want it to check its work without being asked.

In the Code tab of Claude’s desktop app, “Claude can start a dev server and open it in the Browser pane to verify its changes,” for front ends and back ends alike: “Claude can test API endpoints, view server logs, and iterate on issues it finds.” The check is on by default. With auto-verify, “Claude automatically verifies code changes after editing files. It takes screenshots, checks for errors, and confirms changes work before completing its response.” You can watch it in the pane, sign in to sites there, “including popup sign-in flows such as Google OAuth,” and set a custom dev command in .claude/launch.json. It runs on Mac and Windows, and the Linux beta has “live app preview” too.

Price: included in every paid Claude plan, from Pro at $20 a month.

Watch for: “The Browser pane uses a clean browser profile, separate from your personal browser, with none of your saved logins or history,” so for your signed-in sites, Anthropic points you to Claude in Chrome. “In an SSH session, Claude doesn’t have the Browser pane’s tools.” It runs only Claude. And Forever Security’s EMBRACE research in October found a way for a rigged local HTML page in its browser to read other files, so keep the app updated.

The ChatGPT app’s browser, for Codex

Best for: Codex users who want to point at what’s wrong on the page.

In the ChatGPT desktop app, where Codex runs on your computer, the built-in browser “gives you and ChatGPT a shared view of websites and local web apps inside a chat.” Ask for it, or type @Browser, and Codex can “open pages, click, type, inspect rendered state, take screenshots, and verify the result of its work in the page.” Its best trick is Annotate: click an element or drag over an area, write what’s wrong, and Codex works through your comments. Turn on Enable full CDP access under Settings > Browser and it can also “profile JavaScript, inspect console output and network traffic.” For signed-in sites, the browser extension works “in Google Chrome, Microsoft Edge, Brave, Opera, or Vivaldi,” and the app picks “your browser when it needs signed-in browser context, and the built-in browser for localhost.”

Price: Codex comes with every ChatGPT plan, and its pricing lists “Built-in browser previews and comments” on Plus, at $20 a month, Pro, from $100, Business and Enterprise.

Watch for: “Browser isn’t available in Codex CLI or the Codex IDE extension.” “ChatGPT can’t automate file uploads in the built-in browser.” On Linux, the app is “available in preview,” and its Linux page says Computer Use, which is how Codex operates the built-in browser, “is available on macOS and Windows but not yet in the Linux preview.” And on October 29, Forever Security reported that a local HTML page in a rigged project could read any file on the computer through this browser; OpenAI told The Register it is “aware of the issue and working to address it.” Until OpenAI says it’s fixed, don’t open project HTML you didn’t write in it.

Playwright MCP

Best for: a clean, scriptable browser that Claude Code, Codex and almost any other agent can drive.

Microsoft’s Playwright MCP “enables LLMs to interact with web pages through structured accessibility snapshots, bypassing the need for screenshots or visually-tuned models.” One line adds it: claude mcp add playwright npx @playwright/mcp@latest for Claude Code, or codex mcp add playwright npx "@playwright/mcp@latest" for Codex. It drives Chrome, Firefox, WebKit or Edge, opens a visible window by default, “headed by default,” and keeps its sign-ins in a persistent profile between sessions. With --isolated, each session starts clean, and with --extension it can “Connect to a running browser instance (Edge/Chrome only),” your own tabs included. It needs Node.js 18 or newer.

Price: free, and open source under the Apache 2.0 license.

Watch for: “A persistent profile can only be used by one browser instance at a time, so concurrent MCP clients sharing the same workspace will conflict,” so give parallel agents --isolated or their own profile. And Microsoft’s own README says that with a coding agent, “you might benefit from using the CLI+SKILLS instead.”

Playwright CLI

Best for: coding agents that need their context window for the code, not the page.

Playwright CLI is Microsoft’s answer to that note: “If you are using coding agents, that is the best fit.” Instead of loading an MCP server’s tools into every prompt, the agent runs short commands such as playwright-cli click <ref> and reads back only what it asks for, because CLI calls “avoid loading large tool schemas and verbose accessibility trees into the model context.” Install it with npm install -g @playwright/cli@latest, then playwright-cli install --skills so “Claude Code, GitHub Copilot and others will use the locally installed skills.” playwright-cli show opens a dashboard where you can “see and control all running browser sessions.”

Price: free, under the Apache 2.0 license.

Watch for: “Playwright CLI is headless by default,” so you see nothing unless you pass --headed. It “keeps the browser profile in memory by default,” so cookies are “lost when the browser closes” unless you add --persistent, and “A headless session shuts down on its own after an hour without commands.”

Chrome DevTools MCP

Best for: finding why the page is broken or slow, not just whether it works.

Google’s Chrome DevTools MCP “lets your coding agent (such as Antigravity, Claude, Cursor or Copilot) control and inspect a live Chrome browser.” It records performance traces, can “Analyze network requests, take screenshots and check browser console messages (with source-mapped stack traces),” and automates clicks through Puppeteer. Its setup guide gives claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latest for Claude Code and codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latest for Codex, and it ships as a command-line tool and skills too. It pairs well with Playwright: one to drive, one to diagnose.

Price: free, under the Apache 2.0 license.

Watch for: it “officially supports Google Chrome and Chrome for Testing only.” Google’s usage statistics are “enabled by default,” until you pass --no-usage-statistics, and “Performance tools may send trace URLs to the Google CrUX API” unless you add --no-performance-crux. And it “exposes content of the browser instance to the MCP clients,” so keep personal sign-ins out of the browser it drives.

agent-browser

Best for: a quick, low-overhead browser any agent can drive from the shell.

Vercel Labs’ agent-browser is “a native Rust CLI that gives an AI agent a browser it can drive one command at a time.” agent-browser snapshot returns the page “as compact text with a ref such as @e1 for each element,” the agent clicks a ref, and the browser stays open between commands, “so each step is one short shell call.” Install it with npm install -g agent-browser and agent-browser install, which downloads Chrome for Testing; --auto-connect attaches to your running Chrome instead. Teach your agent with npx skills add vercel-labs/agent-browser or one line in AGENTS.md, and agent-browser mcp serves MCP clients. It also reads the console, errors and performance, and ships for macOS, Linux and Windows.

Price: free, under the Apache 2.0 license.

Watch for: “Refs describe the page at the time of the snapshot. After navigation or a DOM change, take a new snapshot before using refs again.” And “the daemon closes a headless browser and exits after one hour without commands.”

Orca

Best for: running many agents in parallel, each with a browser of its own, for free.

Orca runs “Claude Code, Codex, and any other coding agent in parallel, each in its own worktree,” and “Every Orca worktree has its own browser. It’s a real Chromium window — address bar, history, devtools — embedded in a pane,” its docs say. Agents drive it with the Orca CLI, through orca snapshot, orca click and orca fill: “Same browser you interact with, same tabs.” It can import your cookies from Chrome or Edge in one click, emulate a phone’s viewport, and with Design Mode, “Click any element in Orca’s browser to copy its HTML, styles, and selector, then paste it into your agent’s chat.”

Price: free, and open source under the MIT License, on macOS, Windows and Linux.

Watch for: agents reach the browser through the Orca CLI, which ships with the app but which you register under Settings → General → Orca CLI, and “Agents can install the matching Orca CLI skill” to learn its commands. Google sign-ins don’t come over with your cookies: “Google logins are not imported; sign in to Google directly in Orca.”

Enjoy

Best for: letting every lab’s agent open your site beside the conversation and test it, with nothing to set up.

Enjoy runs Claude Code, Codex, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on Mac, Windows and Linux, on the plans you already have, and every agent gets Enjoy’s own tools, its browser among them, “in every conversation, with nothing to set up.” In that browser, an agent can start your site, open it beside the conversation, read the page, click, fill in forms, scroll and go to another address. On a Mac it can also take screenshots, press keys, and hover and drag. It’s the same page you see, so you can watch it work, and it keeps going when you switch to another conversation. Sign-ins you make in the built-in browser stay after you quit, and the agent can use them. Claude Code in Enjoy also has Claude in Chrome turned on, for your own browser’s sign-ins, and you can open the preview from your phone or another browser.

Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and one computer; each other computer is $9.99 a month, and each teammate you invite needs a subscription of their own. See pricing.

Watch for: “On Windows and Linux, agents that try to take a screenshot are told screenshots are only available on macOS.” Among the things its help says the built-in browser can’t do: “Upload files from your computer into the page,” “Read the contents of frames embedded in a page,” and “Open separate windows or download files.” Sign-in buttons that open a window send you to your usual browser. It’s for an agent checking its work as it goes, not a test runner, so it doesn’t replace Playwright tests you keep. Enjoy isn’t open source, and it’s a monthly subscription on top of your AI plans once the free week ends.

When to stay with what you have

If you use one agent, you may already have a browser. Claude’s desktop app checks Claude Code’s changes in its Browser pane by default, the ChatGPT app gives Codex a browser of its own, and in a terminal, claude --chrome is a flag away on a paid Claude plan. If you’d rather not add anything to your plan, Playwright MCP or Playwright CLI cost nothing and work with both agents, and Chrome DevTools MCP is the one to add when the question is why, not whether.

Whichever you pick, an agent’s look at the page isn’t a test suite. For checks that must pass on every change, ask the agent to write tests you keep. Look further when you run several labs’ agents at once and want each one’s browser beside its conversation: Orca does that for free, and Enjoy does it for six agents with nothing to set up.

How we compared

We read Anthropic’s Claude Code docs on Chrome, the desktop app and its Linux beta, Claude’s pricing, OpenAI’s docs on the ChatGPT app’s browser, its browser extension, its Linux app and Codex’s pricing, the GitHub pages of Playwright MCP, Playwright CLI, Chrome DevTools MCP and agent-browser, Orca’s homepage, browser and CLI docs, and Enjoy’s own help on November 8, 2026, and every one is linked below. We quote each maker’s own words for its limits and leave out what it hasn’t documented. These tools change weekly, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.