browsermcp-plus
browsermcp-plus lets MCP clients automate a connected real browser tab through a suite of browser-control tools.
Navigate to URLs, go back/forward, reload, and manage history.
Inspect pages with accessibility snapshots and text search for element references.
Click, hover, type, fill forms, select dropdown options, drag and drop, and press keys/shortcuts.
Upload local files, handle alert/confirm/prompt dialogs, run JavaScript, scroll, and take screenshots.
Read console logs and wait for time or for text to appear/disappear.
List, open, select, and close browser tabs.
Support multiple agents in parallel, each with its own browser tab.
The browsermcp-plus Chrome extension can be loaded in Arc, letting the server automate the user's real Arc browser profile — clicking, typing, uploading files, running JavaScript, navigating and reading pages in the connected tab.
The browsermcp-plus Chrome extension can be loaded in Brave, so the server can drive the user's real Brave browser profile — clicking, typing, uploading files, running JavaScript, navigating, scrolling and reading pages in the connected tab while staying logged in.
Provides browser_evaluate to run arbitrary JavaScript in the current page, optionally scoped to an element from the accessibility snapshot, and browser_get_console_logs to read the page console.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@browsermcp-plus go to my GitHub notifications and summarize what needs attention"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
browsermcp-plus
Let AI apps drive your own browser — including file uploads, tabs and JavaScript.
browsermcp-plus is an MCP server and an open-source Chrome extension. Claude, Cursor, VS Code, Windsurf and any other MCP client can click, type, upload files and read pages in the tab you connect, using your real browser profile: you stay logged in, nothing runs in the cloud, and sites see a normal browser.
It started as a hardened fork of Browser MCP and stays compatible with its extension. Türkçe README
What's different from Browser MCP
Browser MCP | browsermcp-plus | |
File upload | ✗ | ✓ file inputs and "choose file" buttons, no OS dialog |
| freeze the extension | reported and handled |
Tokens per click on a large page | whole page (~14k on Wikipedia) | short report (~30) |
Forms | one call per field |
|
Run JavaScript, tabs, scrolling | ✗ | ✓ |
Extension source | closed | open source ( |
Server reachable from the network | yes, all interfaces | loopback only |
Any web page can connect to the server | yes | no, origin allowlist |
Several agents at once | each new client kills the previous one | every agent in its own tab, in parallel |
Builds from its own repository, tests, CI | ✗ | ✓ unit + real-browser end-to-end tests |
Related MCP server: Chrome MCP Server
Install
1. Extension — download browsermcp-plus-extension-*.zip from the
latest release
and unzip it, then in Chrome (or Edge, Brave, Arc…):
open
chrome://extensionsand enable Developer mode,click Load unpacked and choose the unzipped folder,
if you have the original Browser MCP extension, disable it.
2. Server — add it to your MCP client. Either run it straight from GitHub (needs Node.js 18+ and git):
{
"mcpServers": {
"browser": {
"command": "npx",
"args": ["-y", "github:Novpix/browsermcp-plus"]
}
}
}or download browsermcp-plus.cjs from the release — a single file with no
dependencies — and point the client at it:
{
"mcpServers": {
"browser": {
"command": "node",
"args": ["/path/to/browsermcp-plus.cjs"]
}
}
}Claude Code: claude mcp add browser -- npx -y github:Novpix/browsermcp-plus
3. Connect — open the tab you want to automate, click the extension icon
(Alt+J) and press Connect. The icon shows ON.
Tools
Tool | Description |
| Open a URL ( |
| History and reload |
| Accessibility snapshot with element refs (optionally one subtree) |
| Find elements by text without loading the whole snapshot |
| Act on an element from the snapshot |
| Type into an element (key by key for masked inputs), optionally pressing Enter |
| Fill many fields — text, checkboxes, radios, dropdowns, sliders, dates — in one call |
| Choose dropdown values |
| Drag one element onto another (incl. HTML5 drag and drop) |
| Keys and shortcuts ( |
| Upload local files through a file input or a "choose file" button |
| Accept or dismiss |
| Run JavaScript in the page, optionally on a snapshot element |
| Scroll by pixels or scroll an element into view |
| Work across tabs |
| Wait for time, or for text to appear/disappear |
| Read the page console |
| PNG of the visible viewport |
Actions (click, type, …) answer with a short report — URL, title, whether a new
page loaded, open dialogs, new tabs — instead of the whole page, which keeps
agents fast and their context small. Pass snapshot: true when you want the
page right away. Clicks on disabled or covered elements fail with an
explanation instead of silently hitting the overlay.
Tools carry MCP annotations (readOnlyHint, destructiveHint) so clients can
skip confirmation for read-only ones. With the original Browser MCP extension
everything except upload, forms, dialogs, evaluate, scroll and tabs works.
Options
--port <number> WebSocket port (default 9009)
--session-name <name> agent name shown in the browser (default: current folder)
--allow-origin <origin...> extra extension origins allowed to connect
--no-takeover don't share the browser with a running server
--kill-existing terminate a non-cooperating process on the port
--request-timeout <ms> timeout for a single browser action (default 30000)
--snapshot-max-chars <n> truncate large snapshots (default 80000, 0 = never)
--action-snapshots include the page snapshot in every action result
--verbose debug logging on stderrSeveral agents at once
Every Claude Code session (or other MCP client) that starts browsermcp-plus is an agent with its own tab, and agents work in parallel:
The tabs you connect are shared: an agent takes a free one, or gets a new tab next to them when all are busy. Tabs opened for an agent are grouped and labelled with its name (the project folder, or
--session-name).An agent cannot select or close another agent's tab; when an agent exits, its tab is freed for the next one.
The first server is the hub the extension connects to; the others join it. If the hub exits, another server takes its place automatically.
The extension popup shows every agent, its tab, what it is doing and its last error. "Stop all" disconnects everything.
Security model
The server listens on
127.0.0.1/::1only.Only the extension origins may connect, so web pages you visit cannot talk to the server or impersonate the extension.
The extension acts only on the tab you connect; Chrome shows its "is debugging this browser" bar while it does.
Page content returned by the tools is untrusted input for the model.
Development
npm install
npm run check # typecheck + unit tests + build
npm run test:e2e # Chromium + extension end to end (single and multi-agent)
npm run test:real # background tabs in a real, visible Chromium (needs a display)The wire protocol is documented in src/protocol.ts, the
extension in extension/README.md.
License
Apache-2.0. See LICENSE and NOTICE — based on Browser MCP, which was adapted from Playwright MCP. Not affiliated with the Browser MCP authors.
Available Tools
25 toolsbrowser_clickADestructive
Click an element. Fails with an explanation if the element is disabled or covered by another element (e.g. a cookie banner).
Inputs: ref (required), element (required), snapshot.
browser_dragCDestructive
Perform drag and drop between two elements
Inputs: endRef (required), startRef (required), endElement (required), startElement (required), +1 more.
browser_evaluateADestructive
Run JavaScript in the page and return the JSON-serialisable result. Pass a function: () => document.title, or (element) => element.value together with a ref. Promises are awaited.
Inputs: function (required), ref, element.
browser_file_uploadADestructive
Upload local files through a file input. ref may be the file input itself or the button that opens the file chooser; no dialog is shown.
Inputs: ref (required), paths (required), element (required), snapshot.
browser_fill_formADestructive
Fill several form fields in one call: text fields, checkboxes, radio buttons, native dropdowns and sliders. Faster and more reliable than separate typing and clicking.
Inputs: fields (required), snapshot.
browser_findARead-only
Search the page snapshot for text (case-insensitive) and return only the matching elements with their refs and position in the page. Much smaller than a full snapshot.
Inputs: text (required).
browser_get_console_logsBRead-only
Get the console logs from the browser
Inputs: no named parameters.
browser_go_backB
Go back to the previous page
Inputs: snapshot.
browser_go_forwardC
Go forward to the next page
Inputs: snapshot.
browser_handle_dialogADestructive
Accept or dismiss the JavaScript dialog (alert, confirm, prompt, beforeunload) that is blocking the page
Inputs: accept (required), snapshot, promptText.
browser_hoverC
Hover over element on page
Inputs: ref (required), element (required), snapshot.
browser_press_keyADestructive
Press a key or shortcut, e.g. Enter, Escape, Tab, ArrowDown, PageDown (scrolls), Control+a or a single character
Inputs: key (required), snapshot.
browser_reloadA
Reload the current page by navigating to its current URL
Inputs: snapshot.
browser_screenshotARead-only
Take a screenshot of the visible part of the current page
Inputs: no named parameters.
browser_scrollA
Scroll the page with the mouse wheel by a number of pixels, or scroll an element from the snapshot into view
Inputs: ref, deltaX, deltaY, element, +1 more.
browser_select_optionCDestructive
Select an option in a dropdown
Inputs: ref (required), values (required), element (required), snapshot.
browser_snapshotARead-only
Capture accessibility snapshot of the current page. Use this for getting references to elements to interact with. Pass a ref to see only that part of the page.
Inputs: ref.
browser_tab_closeADestructive
Close a tab (default: the connected one). Closing the connected tab connects the tab that becomes active.
Inputs: tabId.
browser_tab_listARead-only
List open browser tabs. Your tab is marked connected; tabs used by other agents are marked and cannot be selected.
Inputs: no named parameters.
browser_tab_newA
Open a new tab, optionally at a URL, and make it the connected tab
Inputs: url.
browser_tab_selectA
Work in another tab (ids from browser_tab_list). Tabs used by other agents cannot be selected.
Inputs: tabId (required).
browser_typeADestructive
Type text into an editable element, replacing its content. To fill several fields at once, prefer browser_fill_form.
Inputs: ref (required), text (required), element (required), slowly, +2 more.
browser_waitBRead-only
Wait for a specified time in seconds
Inputs: time (required).
browser_wait_forARead-only
Wait until text appears on (or disappears from) the page's accessibility snapshot, then return the snapshot
Inputs: text, timeout, textGone.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 25 tools
Most tools map to clearly distinct browser actions (navigation, interaction, tabs, waiting, inspection) with little overlap. Pairs like browser_type vs browser_fill_form and browser_wait vs browser_wait_for are differentiated by their descriptions, but the high tool count creates minor potential for confusion.
All 25 tools use a consistent browser_ prefix and snake_case, following a predictable verb or noun pattern (e.g., browser_navigate, browser_tab_close, browser_get_console_logs). No mixed conventions are present.
25 tools is on the heavy side for a browser automation server. While each tool has a purpose, several could be consolidated (e.g., wait/wait_for, type/fill_form), placing the count in the borderline-heavy range per the rubric.
The surface covers core browser lifecycle actions: navigation, tab management, interaction, waiting, screenshots, console logs, JS evaluation, and dialog handling. Minor gaps like direct URL/title access or network/cookie management exist but are workaroundable via browser_evaluate.
Maintenance
Related MCP Connectors
Browz hosted MCP (free public beta): let an AI app drive your own Browz browser profiles from anywhere. Connect https://mcp.gobrowz.com/mcp and sign in with your Browz account (OAuth) or an agent token. Each call is relayed to the Browz app on your computer, which must be online with Remote control on and have a hardware security chip (Mac Secure Enclave or TPM 2.0). 17 tools: profiles, tabs, navigate, snapshot, act, wait, screenshot. Cookies, downloads and stored sign-ins stay on your computer.
Browser MCP for logged-in tasks. Uses your Chrome — credentials stay local. Zero-token replay.
Hosted Web Crawler MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceAn MCP server that provides AI assistants with full control over a real browser session via a Chrome extension, supporting 36 tools for navigation, data extraction, and DOM manipulation. It bypasses bot detection by utilizing the user's active browser session, including cookies, authentication tokens, and installed extensions.4MIT
- AlicenseNot gradedqualityDmaintenanceAn extension-based MCP server that enables AI assistants to control your browser, leveraging existing sessions and login states for automation and content analysis. It provides over 20 tools for semantic tab search, interactive element manipulation, and network monitoring directly within your daily Chrome environment.MIT
- AlicenseNot gradedqualityNot gradedmaintenanceAn extension-based MCP server that enables AI assistants to control your existing Chrome browser, leveraging your active login states and settings for automation. It provides over 20 tools for tasks like semantic tab search, screen capture, network monitoring, and direct element interaction.-
- AlicenseBqualityDmaintenanceAn MCP server that provides AI models with full browser automation capabilities through Chrome. It enables navigation, interaction, screenshots, and complete DevTools access by bridging AI clients with a companion Chrome extension.993 npm3Apache 2.0