browser_login tool, or add "env": {"BROWSER_USE_API_KEY": "bu_..."} next to "args". A client launched from a desktop app may not inherit your shell’s exports; set the key in its config if needed. The stored key from browser-use login works too. Keep the exact ==1.0.6 pin: browser-use moves fast, and a new release can break your setup.
Tools
These are the same helpers as the CLI. The client’s model writes the code; there is no MCP tool that runs the full browser-use
Agent.
Local or cloud
- Local:
browser_execdrives the Chrome you have open, with your logins. When Chrome cannot be used (no display, remote debugging off or no usable browser), the server starts its own temporary Chromium, headless without a display, for the rest of the process. It can install Playwright’s Chromium on a headless machine if none is available. The first attached call names the local browser in a separate text block. - Cloud: call
start_remote_daemon()inbrowser_exec(afterbrowser_loginif you are not signed in). Every later call drives that cloud browser. It lives 30 minutes unless you passtimeout=<minutes>(up to 240), and the server stops it when it exits.
BH_OPEN_LIVE_URL=1 is set.
browser_exec runs Python in the server process, with access to its environment and files. http_get() runs from that machine; it does not use the cloud browser’s proxy.
Migrating from 0.x —mcp
The 0.xbrowser-use --mcp server ran the removed engine. --mcp now prints a deprecation notice and serves the same four tools on its own temporary Chromium. Old tool names (browser_navigate, browser_click, retry_with_browser_use_agent, …) return an error that points to browser_exec. It keeps the domain restrictions from the 0.x default browser profile and BROWSER_USE_ALLOWED_DOMAINS; with either set, starting a cloud browser is disabled. A configured local proxy fails instead of being ignored. Switch clients to --cli-mcp.