AugmentClaude

Browser Use

Automate tasks in your existing Chrome browser without creating new profiles.

Installation

  1. Make sure Claude is on your device and in your terminal.

    Skills load from ~/.claude/skills/ when Claude Code starts up — so you need it on your machine first. If you don't have it yet, install it once with the command below, then run claude in any terminal to verify.

    One-time setup
    npm i -g @anthropic-ai/claude-code

    Already have it? Skip ahead.

  2. Paste into Claude Code or into your terminal.

    This copies the whole skill folder into ~/.claude/skills/browser-use-steipete/ — the SKILL.md plus any scripts, reference docs, or templates the skill ships with. Safe default: works for every skill.

    Faster alternative (instruction-only skills)

    Skips the clone and grabs only the SKILL.md file. Don't use this if the skill ships Python scripts, reference markdowns, or asset templates — they won't be downloaded and the skill will fail when it tries to load them.

    Quick install (SKILL.md only)
    Sign up to copy
  3. Restart Claude Code.

    Quit and reopen Claude Code (or any other agent that loads from ~/.claude/skills/). New skills are picked up on startup.

  4. Just ask Claude.

    Skills auto-activate when your request matches the skill's description — no slash command needed. Trigger phrases live in the skill's own frontmatter; you can read them in the “What this skill does” section above.

Prefer to read the source first? Open on GitHub.

When Claude uses it

Chrome DevTools MCP automation: existing Chrome tabs, no AppleScript.

What this skill does

Browser Use

Use this for browser tasks against the existing Chrome session.

Config repair details live in mcporter-config.md.

Hard rule: reattach to the existing Chrome profile only. Use this target:

mcporter call chrome-devtools.<tool>

Most login-heavy sites fail in isolated profiles because fresh sessions trigger captcha, device checks, or missing SSO/extension state. Strongly prefer the existing Chrome profile for any website that needs login.

Never use chrome-isolated, Playwright, Puppeteer, the Codex in-app browser, AppleScript, osascript, GUI scripting, or macOS open for browser control unless the user explicitly asks for an isolated/new browser.

Screenshot/live UI bugs require this existing-Chrome path. curl, source inspection, Worker smoke tests, or local Playwright are supporting proof only; do not treat them as equivalent when the user showed a rendered browser problem or the page may depend on login/profile state.

Check MCP

mcporter list chrome-devtools --schema
mcporter call chrome-devtools.list_pages --args '{}' --output text

list_pages must show the user's real open tabs. If it shows a blank/default isolated Chrome, stop and say reattach failed.

If the call appears to hang while Chrome shows an auth/attach/update prompt, handle the attach alert before falling back. Prefer Peekaboo to press an explicit Chrome Allow button when visible; otherwise wait for the human. Do not restart daemons or kill MCP processes just because the first output is slow.

Tested attach-prompt recovery:

PB="${PEEKABOO_BIN:-$HOME/bin/peekaboo}"
[ -x "$PB" ] || PB="$(command -v peekaboo)"
"$PB" permissions status --json
"$PB" see --app frontmost --path /tmp/chrome-attach.png --json --annotate
# If the UI shows Chrome "Allow remote debugging?", click only the visible Allow button.
"$PB" click --coords <allow_x>,<allow_y> --json
mcporter call chrome-devtools.list_pages --args '{}' --output text

Use coordinates from the current Peekaboo snapshot, not stale notes. Success means list_pages returns the user's real Chrome tabs.

Attach-alert rule: when the current snapshot clearly shows Chrome asking to allow DevTools/MCP/browser automation attachment, click the visible Allow button once, then rerun list_pages. If the button is not visible or the prompt is ambiguous, stop and ask; do not silently switch to Playwright/Puppeteer.

If list_pages fails with DevToolsActivePort, ask the user to restart Chrome or the DevTools bridge, then retry once:

mcporter daemon restart
mcporter call chrome-devtools.list_pages --args '{}' --output text

If it still fails, stop and say Chrome DevTools MCP is unavailable. Do not use AppleScript.

Avoid noisy recovery loops. Repeated MCP/browser restarts can trigger reconnect/login prompts and alerts. Try once, then pause and choose a quieter path.

Typical Flow

# pick the page id from list_pages
mcporter call chrome-devtools.select_page --args '{"pageId":9}' --output text

# inspect page
mcporter call chrome-devtools.take_snapshot --args '{}' --output text

# navigate selected page
mcporter call chrome-devtools.navigate_page --args '{"url":"https://example.com"}' --output text

# click an element uid from the latest snapshot
mcporter call chrome-devtools.click --args '{"uid":"1_38","includeSnapshot":true}' --output text

# type/fill
mcporter call chrome-devtools.fill --args '{"uid":"1_13","value":"text","includeSnapshot":true}' --output text

# run JS, keep secrets out of output
mcporter call chrome-devtools.evaluate_script --args '{"function":"() => document.title"}' --output json

Use take_snapshot before actions and use current uid values only. Avoid take_screenshot unless visual layout matters.

Live UI Proof

For screenshot regressions, deployed dashboard checks, or anything where the rendered browser is the bug:

mcporter call chrome-devtools.list_pages --args '{}' --output text
mcporter call chrome-devtools.select_page --args '{"pageId":9}' --output text
mcporter call chrome-devtools.navigate_page --args '{"url":"https://example.com"}' --output text
mcporter call chrome-devtools.take_snapshot --args '{}' --output text
mcporter call chrome-devtools.evaluate_script --args '{"function":"() => document.body.innerText"}' --output json

Use the existing logged-in/profile-bearing tab set. If browser automation is unavailable, report that as a verification gap instead of substituting isolated browser tooling.

Secret Handling

Never print tokens/passwords from page DOM, network logs, or inputs. For token checks, return shape only: present/absent, length, status code, account/org name.

Related skills