Browser Use Cloud Skill: 112K Stars and Counting — Is This the Documentation Layer Claude Code Was Missing?

I keep a close eye on the SkillsMP trending board, and this week one skill jumped out: cloud by browser-use, sitting at 112,056 stars with a "breakout" trend tag and 1,692 stars gained in the last seven days alone. That's not a typo. For reference, that's more than most popular open-source projects accumulate in a lifetime.

But here's the thing I had to remind myself of: the stars are for the parent browser-use repository. The skill itself is a documentation reference layered on top. Still, the momentum is real, and the team behind one of the most-used browser-automation libraries for LLMs clearly cares enough to maintain a Claude Code companion. So I pulled it apart.

What This Skill Actually Does

Let's get this out of the way: this is not a tool skill. It's not going to spin up a browser session for you or hit an API endpoint. The allowed-tools: Read line in the manifest is a dead giveaway. What cloud does is act as a curated documentation router — when you ask Claude Code about Browser Use Cloud's hosted API, the skill decides which of the bundled reference files to load so the model answers with the right context.

That distinction matters because Browser Use has had a confusing version history. There are v2, v3, and v4 of their cloud REST API. There are three flavors of Python SDK import path. There's a TypeScript SDK with parallel versioning. There are separate concerns: CDP WebSocket connections, browser profiles, profile sync, stealth browsers, residential proxies, CAPTCHA handling, webhooks, workspaces, a skills marketplace of their own, live URL streaming, and pricing. If you've ever tried to grep through their docs site mid-task, you know the pain.

This skill is the antidote. It doesn't teach the model anything new — it just points the model at the right 200 lines of context for the question being asked.

Why It Matters

The gap it fills is the same gap every rapidly-evolving SaaS API fills: docs lag behind reality, and LLM context windows hate ambiguity. Without a routing skill, if I asked Claude Code "how do I run a stealth browser session with a US residential proxy using v4?", the model would either hallucinate an endpoint, confuse v3 and v4 syntax, or pull from stale v2 examples.

The skill's description is explicit about this — it lists dozens of trigger phrases, from "X-Browser-Use-API-Key authentication" to "liveUrl streaming" to "1Password vault integration." That's not an accident. The author is trying to maximize routing precision so the wrong reference file never loads. Smart.

What I appreciate is the explicit negative trigger: "Do NOT use this for the open-source Python library (Agent, Browser, Tools config) — use the open-source skill instead." This is the kind of clarity that prevents skill collisions. If you've installed both cloud and the open-source skill, Claude Code needs to know which one to consult, and the description does that work.

Key Capabilities

Five highlights I want to call out from the SKILL.md:

1. Version-aware routing. Three reference files cover v2 (api-v2.md with 30 endpoints and schemas), v3 (api-v3.md with sessions/messages/files/workspaces), and v4 (api-v4.md for current setup, first run, and workspaces). The skill knows which to load based on your question.

2. CDP escape hatch. The browser-api.md reference covers connecting Playwright, Puppeteer, or Selenium directly to Browser Use's cloud browsers via CDP WebSocket. The endpoint format is documented: wss://connect.browser-use.com?apiKey=KEY&proxyCountryCode=us. If you're already running an automation stack locally and just want a hosted browser, this is the integration path.

3. Integration patterns. Three guides in references/guides/ cover chat UI with live browser view, using browser-use as a subagent (task in, result out), and adding browser-use tools to an existing agent framework. These are the patterns that actually unblock production use cases — the difference between "I played with it for 20 minutes" and "I shipped a feature."

4. Critical operational notes baked in. The skill surfaces things that would otherwise burn you: "Always stop a browser explicitly; closing CDP does not stop billing." That single line probably saves a few hundred dollars per month for someone who didn't read the fine print. Auth header format, env var name (BROWSER_USE_API_KEY), SDK install commands, and the three import paths per language are all in the manifest.

5. Coverage of the boring but essential stuff. Proxies, webhooks, workspaces, MCP, parallel runs, geo-scraping, pricing — these are the things that break demos in production and force you to read six pages of docs. Having them indexed under one skill description is genuinely useful.

Who Should Install This

Install if: - You're building anything on Browser Use Cloud (hosted API, SDK, or CDP-direct). - You're evaluating Browser Use vs. self-hosted Playwright/Puppeteer and want accurate version-specific answers. - You maintain an integration with v2 or v3 and need the model to consistently reach for legacy endpoints instead of bleeding-edge v4. - You're wiring browser-use into Claude Code as a subagent and need the integration patterns indexed.

Skip if: - You're only using the open-source browser-use Python library locally — the manifest explicitly redirects that to a sibling skill. - You don't use Browser Use at all. This won't help you with Playwright, Selenium, or any non-browser-use tool. - You're working completely offline. The skill references live endpoints and external docs that won't help in an air-gapped environment.

How to Install

Clone or download the cloud directory from the browser-use/browser-use repo's skills/ folder and drop it into your Claude Code skills path:

# User-level install (all projects)
git clone https://github.com/browser-use/browser-use.git /tmp/bu-cloud
cp -r /tmp/bu-cloud/skills/cloud ~/.claude/skills/

# Project-level install (this repo only)
cp -r /tmp/bu-cloud/skills/cloud .claude/skills/

Restart Claude Code, and the skill description will register automatically. You can verify by asking "How do I authenticate to Browser Use Cloud?" — you should get a v4-specific answer instead of a generic API key explanation.

Concerns and Limitations

A few honest reservations:

First, it's a passive skill. No allowed-tools beyond Read means it doesn't do anything on its own. If you're looking for a skill that executes browser tasks, this isn't it — you'd want a wrapper that calls the SDK. This is documentation routing, full stop.

Second, version drift is inevitable. Browser Use is iterating fast. v4 is current today, but if v5 ships and the reference files lag, the skill will route you to outdated endpoints. The breakout star count suggests the maintainers are responsive, but it's a thing to watch.

Third, the description is enormous. Trigger phrase lists this long work, but they also bloat every system prompt where the skill is loaded. If you're running tight on context, that matters. For Browser Use users it's worth it; for someone who never touches the tool, it's noise.

Fourth, I couldn't see the bundled reference files from the metadata alone. The SKILL.md indexes them but doesn't include them inline. You'll want to crack open references/api-v4.md, references/features.md, and the integration guides to confirm they match your needs before committing.

Verdict

For anyone working with Browser Use Cloud — whether that's the hosted REST API, the SDK, or a direct CDP connection — this skill is a no-brainer install. It collapses a sprawling, fast-moving documentation surface into a single, well-routed reference, and the manifest shows real thought put into trigger precision and operational gotchas (especially the billing warning).

If you're not using Browser Use Cloud, skip it. There's nothing general here.

The broader lesson is that documentation-routing skills are an underrated category. When a vendor maintains their own Claude Code companion with this level of care, it dramatically shortens the gap between "I have a question" and "I have a working answer." I'd love to see more SaaS vendors ship skills like this.

Install it. It's small, it's targeted, and it'll pay for itself the first time you ask a v3-vs-v4 question mid-task.

→ View on SkillsMP → View on GitHub