Claude and your web browser
Claude can read web pages on its own; that is what the research above uses. Clicking around a site as you, signed in to your accounts, is a separate tool: the Claude in Chrome extension. What works depends on the platform:
- macOS: Claude Code's Chrome integration is documented for macOS. Install the
extension, then start Claude with
claude --chromeor run/chrome. - Windows: Claude Code's documentation says plainly that its Chrome connection "isn't supported in Windows Subsystem for Linux (WSL)", so Claude Code in Ubuntu cannot drive your Windows Chrome. The two options below cover it.
- Native Linux: not claimed here. Use Playwright.
- Playwright works on every platform; it is described after this list.
On Windows, install Claude in Chrome in Chrome on Windows and sign in with your Claude account, and use it on its own. It opens as a side panel, sees the page you are on, and can click, type, and fill in forms in your signed-in sites. It asks before acting on a new site; read what it is about to do before you approve, especially on anything involving money or messages.
Playwright, on any platform, lets Claude Code use a browser of its own, for testing an app
you are building or pulling data from public pages. The playwright part of the installer sets it up:
Playwright's MCP server, an add-on that gives Claude browser tools, driving the Chrome
(or Chromium) installed during setup; it needs a browser present. Ask "open localhost:3000
and check the sign-up form works." This browser is separate from your everyday one, so it
is signed in to nothing. When there is no display to draw a window on, as under WSL
without Linux window support, the installer runs it headless: out of sight, with Claude
still reading and clicking it. claude mcp list shows it, and /mcp inside Claude shows
its tools.
Several conversations at once
One Claude conversation can keep working while you start another. Press the left arrow (←) on an empty prompt and the conversation moves to the background and keeps going. You land in agent view, a list of every background conversation and what each is doing:
- Working: Claude is busy.
- Needs input (yellow): it is waiting on you, for an answer or a permission.
- Idle or Completed: done for now, ready for your next message.
From agent view:
- Type a message at the bottom and press Enter to start a new conversation with it.
- ↑ and ↓ move between conversations; Enter or → opens one; ← on an empty prompt leaves it running and takes you back to the list.
- Space peeks at a conversation and lets you reply without opening it.
- Ctrl+R renames one, so you can tell "resume draft" from "Spanish lesson 3".
- Ctrl+X stops one; press it again within two seconds to delete it.
- Esc goes back to where you came from.
claude agents opens agent view straight from the shell, and /bg followed by an
instruction backgrounds the current conversation with one last task ("/bg finish the
cover letter and save it"). A background conversation keeps running after you close the
window. A plain claude conversation that you never backgrounded ends with its window.
Restarting the PC stops them all, but none is lost: open one again from agent view, or
use /resume for anything older.
Good uses: a long research task in the background while you work on something else, or
one conversation per project (the job search, a course, an app) that you come back to.
Two conversations editing the same files at once can trip over each other, so give each
its own folder. Agent view is a preview feature, so its keys may change; press ? in it
for the current list.
WezTerm tabs are the other way to run conversations side by side: Ctrl+Shift+T (Cmd+T on a Mac) opens a new tab, and each tab is its own terminal window. Tabs end when you close them; background conversations don't.