Connect an AI assistant with MCP
The Browser.lol MCP server lets your assistant use a browser running in the cloud. Here is how to connect a client, use the tools, and manage sessions and access.
Add the server to your client
Connect Claude Code
Install Claude Code and sign in. Use Bash or Zsh (macOS, Linux or WSL) for the command below.
Add the connection
Run the command in your terminal. It adds Browser.lol to your personal Claude Code settings for all projects.
Run in your terminal
claude mcp add --scope user --transport http browserlol https://api.browser.lol/mcpSign in and allow access
In Claude Code, type /mcp, choose browserlol and sign in on the Browser.lol page that opens.
/mcpbrowserlolAuthenticate
Try it in a new chat
Check in /mcp that browserlol is connected. Then send the task below and approve the browser tools when asked.
/mcp
Open example.com in a Browser.lol session and tell me what the page says.
What you need
MCP (Model Context Protocol) lets an assistant use external tools. Connect yours to Browser.lol for a cloud browser it operates like a person: real mouse and keyboard input with human timing, so the sites it visits see a visitor, not automation. It can open and read pages, interact with them, upload files, and capture the whole screen. Available browser images and exit locations depend on your account.
The browser runs on our infrastructure, so you do not need to install a browser driver. Your MCP client still runs where you set it up. We host the session and offer a live view; your assistant's model stays with its client.
What you need
- Endpoint
- https://api.browser.lol/mcp
- Transport
- Streamable HTTP, stateless per request
- Authentication
- OAuth sign-in in your browser, or an API key in a header for automation
- Plan
- An account with programmatic access, included with Premium
- Browsers
- Images marked for programmatic use and available to your account; Chrome by default
The tools your assistant gets
Tools named browser_* operate the browser like a person: every click, keystroke and scroll is real mouse and keyboard input with human timing, so sites see a visitor, not automation. An assistant can start with browser_navigate, which can open a session and return an accessibility snapshot; its element references work with the page tools. browser_screenshot always captures the whole screen, and its pixels are the x/y coordinates that browser_click, browser_hover, browser_drag and browser_scroll take.
Sessions and account
Start sessions with a screen size, language and keyboard layout; list, rename and close them; change the exit location; read and update your account; read the temporary mail inbox for sign-up links and codes; and ask the person watching for help with a login or captcha.
- browser_session_create
- browser_session_list
- browser_session_update
- browser_session_close
- browser_account
- browser_account_update
- browser_mail
- browser_ask_user
Navigation
Open a URL, go back or forward, reload, and work in several tabs.
- browser_navigate
- browser_tabs
Reading the page
Read accessibility snapshots with dialogs listed first, search page text, wait for text, a URL or a delay, inspect console and network activity, read cookies and storage, and capture the whole screen, including the browser toolbar, menus and native dialogs.
- browser_snapshot
- browser_find
- browser_screenshot
- browser_wait_for
- browser_devtools
Interaction
Click, type, fill forms, select options and scroll, by element reference or by the screen coordinates from browser_screenshot. Typing without a reference goes wherever the keyboard focus is, including the address bar; key presses also answer a JavaScript dialog with Enter or Escape.
- browser_click
- browser_type
- browser_fill_form
- browser_select_option
- browser_press_key
- browser_scroll
- browser_hover
- browser_drag
Files
Upload a file through a page input or retrieve a file downloaded by the session.
- browser_file_upload
- browser_downloads
How the sessions behave
- Page tools can start a session. If your account has no running session, a page tool can create one. If several sessions are running, specify a session id so the server knows which one to use.
- Sessions have time limits. Sessions can end after their plan's idle or maximum runtime limits. You can close one yourself to free its slot.
- People and agents share sessions. Your assistant can list and use sessions started in the app, including saved-profile sessions. You can watch and manage sessions it starts. Changes in a saved-profile session may persist.
- Access stays within your account. Tools can access sessions on the authenticated account, subject to that account's permissions.
- Programmatic access is required. The account needs the programmatic access feature. Without it, the MCP access gate returns HTTP 403. Premium includes the feature.
Security
- Connected apps and API keys act on your account like a password. Disconnect an app or revoke a key on the Agents page as soon as you no longer use it.
- Send an API key in an Authorization or x-api-key header. Keys in the URL are not accepted, because addresses end up in logs and browser histories.
- Pages load in remote browser containers. An agent can also use a session with a saved profile, where its changes may persist.
- Control is automatic: your assistant's actions drive the browser while you watch in the live viewer. Take over with the button there: while you are in control, the assistant pauses, its actions wait and report user_in_control with a time to retry, and reading still works. It continues once you hand control back.
Explore programmatic access