Browser Control

Let Klaus open pages, click links, fill forms, and take screenshots to read pages on its own. Two ways to run it:

• Built-in browser: the browsing view that ships with Klaus, ready to use out of the box, but without your usual sign-in state.
• Chrome bridge: takes over your own Chrome once the Chrome extension is installed, so sites you are already signed in to just work.

Browser control is now a built-in plugin, and its toggle and configuration surface live on that plugin's card (it used to be a native capability card on the MCP page — see MCP and connectors). The web search capability that uses Bing browser search by default also runs on this browser stack, so built-in search requires this plugin to stay enabled; see "Web access" in Preferences.

Installing the Chrome extension

Taking over your own Chrome requires installing an extension and connecting it to the local message relay (native host). The card offers two paths, plus diagnostics:

• One-click install prep: combines four things into a single click — deploying the extension files, copying its local path, opening the Chrome extensions page, and revealing the extension directory in Finder. If you want to see every step, use the step-by-step section below, but the one-click path is not a required detour.
• Zero-gesture install via system policy (macOS only): writes Chrome's ExtensionInstallForcelist policy directly so the extension installs itself, with no gesture from you at all. This path detects whether the machine is managed by MDM — on a managed machine the policy comes from IT, so Klaus tells you why this route will not work and falls back to manual installation instead of failing silently. Both installing and uninstalling this way require your explicit consent.
• Diagnostics: reports whether the extension is connected, whether the native host is installed, which Chrome instance is currently connected (with its version), and which item to check when a click seems to do nothing.

Page permissions and tab references

When a page in the built-in browser requests a permission such as location, notifications, or the camera, Klaus does not show the system's native prompt; it shows its own "Browser permission" dialog naming the site and the permission, and you choose whether to allow it.

To have Klaus work on a page you already have open, type @ in the composer and pick it under "Tabs". The reference points at exactly that browser and tab, and works in queued messages too, so Klaus does not have to guess which page you mean.

Configurable options

These are not laid out as controls on a settings page, but they can all be changed — just ask Klaus to change them and it will call the browser's own control surface:

• Confirmation policy (which actions require asking you), screenshot policy, and the allowlist of sites it may access (add or remove domains).
• Browsing data can be cleared (cookies, cache, or everything). This step is irreversible, so it requires your explicit consent before it runs.