Docs
Tools
Browser MCP exposes 40 tools to the connected agent. Page and tab tools act on the session's own tabs, inside its own tab group: the tab it last used, or a tab or frame it targets explicitly. browser_fetch sends its request from the extension in the background, without a tab, and browser_about and browser_provide_feedback are answered by the local server. The browser work runs in your Chrome through the Browser MCP extension - no headless browser, no Playwright binary.
40tools, in 9 groups
5Navigation & Content14Interaction6Tabs & Frames5Data & Storage2Files3Network1CAPTCHA1Human-in-the-Loop3Meta & Recovery
You never call these by name. You describe what you want, and the client picks:
You: Fill in the address form with my work address and submit it.
Claude: [browser_fill, browser_fill, browser_fill, browser_click]
Submitted. The page now shows "Address saved".The list below is what the client has to choose from.
Navigation & Content - 5 tools
| Tool | Description |
|---|---|
browser_navigate | Navigate the active tab to a URL; reuses the current tab unless new_tab=true. |
browser_get_page_content | Return the current page's content as text or html. |
browser_screenshot | Screenshot the visible viewport; returns base64 PNG or saves to a given path. |
browser_execute_script | Run arbitrary JavaScript in the page context and return the result. |
browser_extract_list | Read the rows of a long or virtualised list by scrolling its container until nothing new appears - mail lists, invoice tables, transaction histories. Those UIs keep only a handful of rows in the DOM, so a single page read returns a sliver and looks complete. In the active tab it scrolls with real wheel events, also at the bottom, so feeds that load on wheel (Threads, X) load; a hidden page (a background tab) gets no scroll events, so there a list that loads or renders while scrolling stops at what is already there, and reached_end is false. reached_end is true only when the list stood at its bottom with nothing new for several rounds. Rows with the same text come back once, so compare the count with what the page says. |
Interaction - 14 tools
| Tool | Description |
|---|---|
browser_click | Click an element via CSS or text selector (text=Submit, button:text(Next)); auto-scrolls into view, uses real mouse events. A text selector looks inside an open modal dialog first. If the mouse would not reach the target at its center (something in front of it, a target that does not take clicks there, or a different control inside it - a web component's own button that covers it is not one, and nor is a box with a checkbox, switch or radio role right next to a hidden field and the only one in the nearest label around it, a label of that field - a box inside a closed shadow root cannot be seen and is not counted), the answer is covered; if something gets in the way only after the press, the answer is covered or maybe_landed, never a plain yes. If the page cannot be checked, nothing is sent and the answer says so. |
browser_double_click | Double-click an element - for editors and grids that open on double-click rather than single. Answers covered when the mouse would not reach it. |
browser_right_click | Right-click an element to open its context menu. Answers covered when the mouse would not reach it. |
browser_click_xy | Click at absolute viewport coordinates. Last resort for canvas, maps, and custom-rendered UI where no element can be selected. |
browser_fill | Fill a form input via CSS or text selector; works on CSP-strict sites via the Chrome Debugger API. With a text selector, nothing is typed (covered) when the mouse would not reach the field. |
browser_press_key | Send a keyboard key press (Enter, Tab, Escape, arrows, letters...) with optional ctrl/alt/shift/meta modifiers. |
browser_scroll | Scroll to a matched element, or by a pixel offset. |
browser_wait | Wait for an element matching a CSS or text selector to appear. |
browser_hover | Hover an element to trigger tooltips, dropdowns, or hover states. Answers covered when the mouse would not reach it. |
browser_select_option | Select an option from a native <select> or a custom dropdown (Angular Material, React Select, etc.). A custom trigger or option the mouse would not reach is not clicked (covered). |
browser_set_combobox | Drive an autocomplete/combobox: click, type filter query, wait for the listbox, click the option(s); supports multi-select chips. |
browser_set_date | Set a date input robustly - native value-set, masked-text typing, or calendar-picker navigation (MUI/AntD/react-datepicker/Lexical), with read-back verification. |
browser_dismiss_overlays | Bulk-dismiss popups, modals, tooltips, and banners via heuristics on close affordances (aria-label, "Skip"/"Ikke nu"/"Got it", × button). |
browser_handle_dialog | Accept or dismiss a native alert()/confirm()/prompt() dialog; call before the action that triggers it. |
Tabs & Frames - 6 tools
| Tool | Description |
|---|---|
browser_list_tabs | List the current session's own open tabs (URL + title) - each concurrent session has its own tab group and can't see another session's tabs. |
browser_switch_tab | Activate a specific tab by ID. |
browser_close_tab | Close a tab by ID (session-owned tabs only). |
browser_get_new_tab | Return the most recently opened tab - useful after a link opens a new tab or an OAuth popup. |
browser_list_frames | List all iframes on the current page with URL and index. |
browser_select_frame | Execute JavaScript inside a specific iframe, targeted by index. |
Data & Storage - 5 tools
| Tool | Description |
|---|---|
browser_get_cookies | Get cookies for a site this session has open. |
browser_set_cookies | Set one or more cookies for a domain (single cookie or a cookies[] batch). |
browser_get_local_storage | Read localStorage from the current page - a single key or all of it. |
browser_set_local_storage | Write a localStorage key/value pair on the current page. |
browser_console_logs | Return recent console.log/warn/error messages captured from the page. |
Files - 2 tools
| Tool | Description |
|---|---|
browser_upload_file | Upload file(s) to an <input type="file"> via the Chrome Debugger API - no OS file dialog needed. Files must be regular files inside the folder your agent's server runs in; anything else is refused. Chrome only hands a file to the page when "Allow access to file URLs" is on for the Agent360 Browser MCP extension (off by default for Chrome Web Store installs); the switch also lets the extension open any local file as a page, so turning it on is your call. A "Not allowed" error from the handoff is reported as file-access-off unless Chrome confirms that file access is on. |
browser_drop_file | Upload into a drag-drop zone by locating a hidden file input in its subtree, then up to 2 parent levels, and as a last resort the first file input on the page (which can belong to a different upload field); with no input at all it intercepts the native file chooser. Use when browser_upload_file finds no input. Needs the same switch as browser_upload_file. Files must be regular files inside the folder your agent's server runs in; anything else is refused. |
Network - 3 tools
| Tool | Description |
|---|---|
browser_fetch | Make an HTTPS request from the extension background - not subject to the page's CORS or CSP. Your browser cookies are not sent and the answer's cookies are not stored, so pass a token in headers. Plain http works only to 127.0.0.1. |
browser_wait_for_network | Wait for a network request matching a URL substring to complete, via Chrome DevTools Protocol. |
browser_extract_token | Opens a known provider's API page and returns instructions for finding or creating the credentials there, for 9 providers (Stripe, HubSpot, Slack, Shopify, Mailchimp, Pipedrive, Calendly, Google, LinkedIn). It reads nothing itself; the agent then reads the page with browser_get_page_content. For HubSpot, Slack, Shopify and LinkedIn the page is a settings or app list on the way to the token, not the token page itself. Any other provider answers Unknown provider; the agent can still navigate and read the page itself. |
CAPTCHA - 1 tool
| Tool | Description |
|---|---|
browser_solve_captcha | Detect reCAPTCHA v2/v3, hCaptcha, Cloudflare Turnstile or FunCaptcha and work through it one step per call: try the reCAPTCHA checkbox, click the reCAPTCHA image-challenge cells the agent picks from a screenshot it takes first, or hand it to you through browser_ask_user. |
Human-in-the-Loop - 1 tool
| Tool | Description |
|---|---|
browser_ask_user | Ask the user to act or answer: without fields, a small card they can drag aside while they act on the page (log in, solve a CAPTCHA); with fields, a dialog over the page. It is drawn again after a navigation in the tab, but a prompt with fields ends if the tab moves to another origin (another domain, subdomain or port) or, on a local file, to another file. The page can see the keystrokes, so ask only for secrets that belong to that page. It needs a page that loaded: on Chrome's error page or about:blank it fails at once and shows nothing. It makes the tab active and restores a minimized window, but does not bring Chrome in front of other apps. |
Frequently asked questions
How many tools are there?
- A check on every release counts the tool definitions in the server and compares them with every "N tools" claim on the site and in the README.
What does it mean when a tool answers "unverified"? That the action was sent and the effect could not be read back. It is not a failure and not a success. Repeating the action blindly is the one thing you should not do - read the page instead. Nine tools were changed in 1.29.2 to be able to give this answer rather than a false yes.
Why do some tools need the tab in front? Chrome accepts mouse and keyboard commands for a background tab and silently drops them. Since 1.29.2 the tools measure whether the page actually received the event, so you get an honest failure with the remedy - call browser_switch_tab - instead of a silent one.
Can a tool ask me something in the middle of a run? Yes. browser_ask_user pauses, asks you on your own screen - a 2FA code, a CAPTCHA, a choice only you can make - and carries on in the same tab.
Meta & Recovery - 3 tools
| Tool | Description |
|---|---|
browser_provide_feedback | Self-check plus report in one call. Compares the connected extension against this server and detects more than one Browser MCP extension connected at once, also the same extension connected more than once (for instance from several Chrome profiles or browsers); with BROWSER_MCP_CHECK_NPM=1 it also compares this server against the latest on npm - the three things that explain most "it just stopped working" moments. Returns a verdict, concrete fix steps, and a pre-filled issue link for whatever is genuinely missing. The agent calls it on its own whenever a tool blocks it. |
browser_about | Return Browser MCP info plus pre-filled links for the user to submit a feature wish, share a use-case, or report a bug. |
browser_reattach_debugger | Force-detach and re-attach the Chrome debugger on the current tab. Use when click/fill/press_key start timing out or report a ghost attach while browser_list_tabs still works - faster than reloading the extension. |
Total: 40 tools (5 + 14 + 6 + 5 + 2 + 3 + 1 + 1 + 3), verified against mcp-server/tools.js line-by-line - no invented tools.