The built-in browser: preview your app, browse and capture

A browser the assistant can see

Coding agents are usually blind to the thing they’re building — they write the code but never look at the running page. More AI closes that loop with a real, embedded browser that lives beside your work in Code mode, so you can show the assistant exactly what you see instead of describing it.

It’s a genuine browser: an address bar that doubles as search, back and forward, reload, DevTools, and “open in your default browser” when you’d rather leave. It runs sandboxed in its own session, over http and https only.

Preview your dev server in one click

The browser reads your project’s dev-server setup — a .claude/launch.json config, or the dev script in package.json — and lists the servers it finds. Start one, and the panel opens it at localhost automatically; stop it just as easily. No hunting for the port, no second window.

It’s the fastest way to go from a change to seeing the change: the agent edits the code, you hit preview, and the running app is right there.

Screenshot, annotate, hand it over

See something off? Capture the page as a screenshot, then mark it up right there — pen, lines, boxes, ellipses and text labels — and add it to the chat. The assistant receives it as an image it can actually look at, so “the button is misaligned here” becomes a picture with an arrow instead of a paragraph.

This turns visual bugs into something the agent can act on directly, which is exactly the context a text-only chat can never carry.

Point at a single element

For a precise fix, use the element picker: click an element on the page and it becomes a chip in your message — the exact thing you mean, handed to the assistant without copying selectors by hand. Edit the chip before you send if you want to narrow it further.

It’s still your browser, and it stays local

The browser is driven by you, not silently by the agent: pages you open and screenshots you take are shared with the assistant only when you choose to. It keeps its own session, remembers the last address per code session, and — like the rest of More AI — sends nothing anywhere except the model provider whose key you supplied.

Frequently asked questions

What is the built-in browser for?

It lets you preview your running app and browse the web without leaving More AI, and — crucially — lets the coding agent see the page. You can screenshot, annotate and point at elements, then hand them to the assistant as real visual context.

Can it preview my local dev server?

Yes. It reads your .claude/launch.json (or the dev script in package.json), lists the servers it finds, and opens the one you start at its localhost address in one click.

How does annotating a screenshot help the assistant?

A marked-up screenshot goes into the chat as an image the model can see, so a visual problem — a misaligned button, a broken layout — becomes a picture with your notes on it rather than a description it has to imagine.

Does the agent browse on its own?

The browser is driven by you. Pages you open and screenshots you capture are shared with the assistant only when you add them to the chat, so you stay in control of what it sees.

Is the built-in browser private?

It runs sandboxed in its own session over http/https only, and like the rest of the app it sends nothing anywhere except the model provider whose key you supplied. Nothing you browse leaves your machine on its own.