browser

name

browser

description

Open and drive a REAL Chromium browser pane inside anu's tmux session. Use this whenever you need to actually SEE or interact with a web page — check a localhost dev preview after starting a dev server, verify a deploy looks right, read a docs/page that needs JavaScript, reproduce a UI bug, or fill out / click through a web form. Renders a real page (JS/CSS/Canvas/WebGL) via the kitty graphics protocol, in the same Chromium the human sees, so they watch your actions live. Drive it with the `browser` command or the browser_* MCP tools (open/navigate/screenshot/read/snapshot/eval/click/crop/type).

browser: a real browser pane you can drive

anu can open a real Chromium browser as a tmux pane — not a text browser. The page renders for real (JavaScript, CSS, Canvas, WebGL). You open one beside you and drive it; the human watches it live in the same Chromium.

The same plugin also provides pdfpane, a native in-pane PDF previewer. Use pdfpane <file.pdf> when the user wants a PDF visible inside anu without going through Chromium's PDF plugin. It watches the file and repaints after rebuilds. Inside the pane, use +/- to zoom, 0 for fit-page, w for fit-width, h/j/k/l or arrows to pan, and n/p to change pages.

The one thing to know

browser [url]                 # open a browser pane beside you -> prints its %NN
browser navigate [%pane] <url>     # load a url
browser back|forward|reload [%pane] # history navigation / reload
browser screenshot [%pane] [out]   # save a PNG -> READ that path to SEE the page
browser read [%pane]               # the page's visible text
browser snapshot [%pane]           # JSON: title/url/text + visible controls/visual elements + ids
browser eval [%pane] <js...>       # run JS, get the result (e.g. document.title)
browser click [%pane] --role link --name "Docs"  # strict semantic click
browser click-id [%pane] <id>      # click an id from the latest snapshot
browser fill [%pane] --role textbox --name "Search" --value "query"
browser crop-id [%pane] <id> [out] # crop a visual element/plot/control
browser wait-text [%pane] "Loaded" --timeout 10000
browser type [%pane] <text...>     # type into the focused element
browser downloads [%pane]          # list this pane's download directory/files
browser ls                         # list open browser panes
browser close [%pane]              # close one
pdfpane [--here] file.pdf [page]   # render a PDF directly in an anu pane

[%pane] is the pane id from browser / browser ls. Omit it when only one browser pane is open — the command resolves the sole pane. Every action drives the SAME Chromium the human sees, so they watch it happen.

How you actually use it

Use semantic state first, then pixels when the task is visual:

• snapshot is the fast structured view. It returns visible controls and visual elements with id, role, name, text, selector, and geometry.
• Prefer click-id, fill-id, --selector, or --role ... --name ... over free text. Free text is exact and fails when ambiguous instead of guessing.
• crop-id, crop --selector ..., and screenshot are your eyes for plots, canvases, layout, and anything whose meaning is visual.
• Use eval for page/app state and wait-text|wait-url|wait-selector after navigation or async actions.

# Check a localhost dev preview after starting the dev server
p=$(browser http://localhost:5173 | grep -oE '%[0-9]+')
browser screenshot "$p" /tmp/preview.png    # then Read /tmp/preview.png to SEE it
browser read "$p"                            # or just read the text

# Verify a deploy, then drill in
browser navigate "$p" https://staging.example.com
browser eval "$p" 'document.title'
browser snapshot "$p"                # find ids, roles, names, selectors
browser click "$p" --role link --name 'Pricing'
browser wait-url "$p" /pricing
browser crop "$p" --selector 'main' /tmp/pricing-main.png

Via MCP (if you have the browser_* tools)

Same capabilities as tools: browser_open (returns the pane id), browser_navigate, browser_back, browser_forward, browser_reload, browser_screenshot (returns a PNG path — read it to see the page), browser_read, browser_snapshot, browser_eval, browser_click, browser_fill, browser_crop, browser_wait, browser_type, browser_downloads, browser_list, browser_close. Prefer these when available; they pass arguments without shell quoting.

Notes

• Engine: default is Helium; override with --engine chrome|chromium|<path> or ANU_BROWSER_ENGINE. It's a fresh, isolated, headless Chromium per pane (its own profile) — not the human's logged-in browser.
• PDFs: prefer pdfpane for local PDF preview panes, especially LaTeX watch workflows. It uses macOS PDFKit by default and can use a compatible renderer via ANU_PDF_RENDERER.
• Keybind: the human opens one with prefix + b.
• Requires a kitty-graphics terminal (Ghostty/kitty) and tmux — already true in an anu session. Each pane is isolated and torn down when closed.
• Close panes you opened when you're done (browser close %NN).