UseBrowser

Let Claude

UseBrowser^*

Like, finally.

A browser for agents. Not a browser with agents.

* name subject to change

Please Don’t Stop The Magic

Claude Code gave us the magic to stop thinking about how and start thinking about what. Ideas become action, collapsing days of work into hours.

We’ve taken this magic for granted — until we open a browser.

Then you’re back to doing things by hand. Research Claude would finish in seconds takes you an afternoon. A workflow it automates before breakfast takes you all week. You know how fast this should be. You’ve lived it. But here you are, clicking through pages like it’s 2019, pretending this is fine.

It’s not fine.

It’s unbearable.

It feels stupid.

There’s only one way to fix this.

Let Claude UseBrowser.*

We forked Chromium from source and built a new type of browser — a browser for agents, totally unlocked. Every API. Every hook. Every access point. Agents don’t just visit pages — they have full control. We built this not for fancy demos, but for serious work. And more importantly, because we had no choice. We refused to live in a world where the magic stops at the browser. We refused to accept that moving slow is okay.

Features

Cmd+ICmd+/

Right here, right now

Cmd+I to start working on something immediately — full Claude Code, right on the page. Cmd+/ for a quick thought — open the drawer, ask, get an answer, keep going. No tab switching to claude.ai. No friction. The AI is right here, at the point of intent.

P.S. It’s also a full terminal. No more spawning shells or switching windows — it’s all in the browser.

Controls →

Screenshot or short clip: Full UseBrowser window. A webpage is loaded. The Cmd+I agent palette is visible at the top with a typed prompt. The built-in terminal is open on the right showing Claude Code executing. The Cmd+/ AI Drawer is floating over the page with Claude.ai loaded. All three interaction surfaces visible simultaneously — command palette, terminal, and drawer.

CLAUDE.md/ub:skills

Vibe tune your browser

Every other browser is a black box — you can’t change how it works. This one, you can. Talk to Claude Code, tweak your CLAUDE.md, and the agent changes how it approaches every task. Just describe what you want. Build the tool that serves you best — and do it fast.

Personalization →

Screenshot or short clip: Split view. Left side shows CLAUDE.user.md open in the terminal editor with user preferences visible (save paths, currency, workflow shortcuts). Right side shows Claude responding to a task using those preferences — saving to the custom path, using MYR currency, outputting a table instead of bullets. The before/after behavior difference in one frame.

/ub:record/ub:learn

Watch, learn, repeat

You don’t start from scratch every day — that’s not how progress works. Record a workflow or teach Claude from a conversation — it writes real Playwright scripts on your machine. Every skill becomes a building block for the next one. The more you invest, the more you have. Your effort compounds. That’s the whole point.

Skills →

Screenshot or short clip: The terminal showing /ub:learn in progress. Claude is analyzing a completed conversation, listing the tool calls it found, and generating a Playwright script. The sidebar skills tab is visible on the left with the newly created skill appearing in the library. Show the progression: conversation → analysis → generated code → saved skill.

Remote Session

Stuck? It calls you

We don’t chase perfection — we chase uptime. When the agent gets stuck, the smartest thing it can do is ask for help. So it does. It pings you on Telegram, you solve it from your phone, and the loop keeps going. Zero downtime. The AI doesn’t need to be perfect. It just needs to keep working — together with you.

Workflows →

Composite image: Left — phone screen showing a Telegram notification: “UseBrowser: Stuck on Shopee login CAPTCHA” with a viewer link. Right — the remote viewer in a mobile browser showing the live browser screen with a slider CAPTCHA visible. The two halves tell the story: notification arrives, user opens the viewer, solves the CAPTCHA from their phone.

MCP

Works with any agent — not just Claude Code

Dozens of MCP tools. Connect Claude Code, remote agents, or your own scripts — any AI agent can drive your browser. You don’t need our agent to use our browser.

MCP Overview →