Browser Use

Browser Use — act inside the user's real Chrome

Wisp talks to the Browser Runtime. Shared mode uses the user's daily Chrome via the unpacked extension — every action runs in their real profile: existing cookies, logins, extensions, and normal fingerprint all apply. Workspace mode can launch a separate Chrome profile. Omitting session always uses shared, even when workspace is connected. Pass session: "workspace" only when the user explicitly requests isolation. If Settings → Browser has Open browser automatically enabled (the default) and the shared extension is disconnected, Wisp may start the installed Chrome/Chromium/Edge so the extension can reconnect. That is still the user's profile, not Playwright or Selenium.

Shared is the default; workspace is not a fallback. Google Chrome 137 and later ignore --load-extension, so on a machine with only branded Chrome the workspace window cannot load the Wisp extension at all. browser_setup {"action":"start_workspace"} therefore returns only once the workspace extension has connected, and otherwise closes the window and fails with WORKSPACE_EXTENSION_BLOCKED. On that error: relay the message, do not retry start_workspace, do not claim any workspace page was opened or read, and get the shared session working instead.

Start with browser_setup without an action (an empty action is also a status check). If the task supplies a target URL, pass it as url so a disconnected daily browser opens that page directly; otherwise startup uses its new-tab page. A connected shared browser is reused without opening another window. Startup does not close existing tabs, including pre-existing blank or new-tab pages.

For figures/code extraction use web_scan with mode: "article" then web_save_assets. For an already-logged-in in-browser chat (ChatGPT, Gemini, or Google AI Mode at google.com/search?udm=50) use web_agent_send, web_agent_wait, web_agent_read on that tab.

Every web_scan and web_execute_js call needs the user's approval by design. Do not treat that as a bug to route around.

Before anything: confirm the bridge is live

Call browser_setup. If status is not connected (or live_retrieval is false), relay its steps (load the unpacked extension from extension_path, verbatim) and stop. Do not answer live, latest, current, or URL-specific questions from prior knowledge. Tell the user this turn contains no live web retrieval and wait until the popup shows Connected to Wisp. Only continue from memory if they explicitly ask for a knowledge-only answer. Never invent the path.

Two fields say why a browser that looks connected is not usable — never report a bare "not connected" when either is set:

• refused_connection — something reached the bridge port and Wisp refused it (usually a different extension id, or another loopback bridge holding the port). Its popup can still read Connected to Wisp. Relay refused_connection.explanation.
• update_required / reload_required — a connected extension is older than this build. Call browser_setup {"action":"update_extension"} first. If it returns updated, call browser_setup again and continue only when update_required=false. If it returns manual_reload_required, relay the current and bundled versions plus extension_path verbatim, ask the user to Reload Wisp Real Browser Bridge on chrome://extensions, and wait. Older unpacked extensions cannot accept Wisp's automatic service-worker reload.

One exception: the user says the extension is already installed. Chrome suspends its service worker when idle and reconnects on a one-minute alarm, so disconnected can just be a sleeping worker. Try web_open_tab or web_scan once — a successful call proves the bridge is live — and relay the install steps only if that call fails too.

The loop

1. web_open_tab {url} — open the page (works even with no tab open yet). Waits until the document is complete, then returns the new tab id plus ready. If ready is false, the load timed out — call web_scan before acting.
2. web_scan — read the page (after waiting for document complete). Returns page.text, page.title, page.ready_state, and page.elements[], where each element carries a unique selector, its visible text/aria_label, and a rect [x,y,w,h]. Use these selectors directly — do not guess. If ready is false, scan again; do not click a partial page. Use tabs_only:true first when you are unsure which tab to target; pass switch_tab_id:<id> to pin one.
3. web_execute_js — act, then re-scan to confirm the effect. The extension waits for complete before running the script, and again if the script navigates.

Recipes (web_execute_js script)

script may instead be a JSON command:

Prefer plain JS. Reach for cmd:cdp only when a page blocks synthetic events or you truly need trusted input.