This repository contains a customized Playwright MCP for RoxyBrowser, designed to enable RoxyBrowser users to control RoxyChrome-based browsers through AI. For more details, please refer to: RoxyBrowser Now Supports MCP — AI Can Finally Do the Work for You
Playwright MCP
A Model Context Protocol (MCP) server that provides browser automation capabilities using Playwright. This server enables LLMs to interact with web pages through structured accessibility snapshots, bypassing the need for screenshots or visually-tuned models.
Key Features
- Fast and lightweight. Uses Playwright's accessibility tree, not pixel-based input.
- LLM-friendly. No vision models needed, operates purely on structured data.
- Deterministic tool application. Avoids ambiguity common with screenshot-based approaches.
- Token-optimized. Advanced DOM distillation reduces token usage by up to 99% while maintaining full functionality.
Token Optimization
The server includes an intelligent DOM distillation system that dramatically reduces token consumption without sacrificing LLM capabilities.
Performance:
- 99% token reduction on complex pages (5MB → 2KB)
- 90-95% fewer elements sent to LLM (only interactive/visible elements)
- Faster snapshots (~150ms vs ~200ms)
- Zero impact on LLM's ability to interact with pages
How It Works:
The optimization system uses viewport-aware filtering and element ID mapping:
- Filters elements by visibility (CSS properties, viewport intersection, dimensions)
- Assigns temporary IDs (
data-mcp-id) to interactive elements - Returns simplified HTML instead of verbose ARIA trees
- Tools accept either traditional
refor optimizedmcpIdparameters
Example Output:
Traditional ARIA mode (verbose):
- document
- banner
- navigation
- list
- listitem
- link "Home" [ref=e1]
# ... hundreds more lines
Optimized mode (concise):
<a id="1">Home</a>
<button id="2">Products</button>
<input id="3" type="search" placeholder="Search..." />
<button id="4">Login</button>
Configuration:
Three snapshot modes available via snapshotMode config:
auto(default): Try optimized, fall back to ARIA on errorsoptimized: Always use token-optimized snapshotsaria: Traditional ARIA snapshots (backward compatible)
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["@playwright/mcp@latest"],
"env": {
"PLAYWRIGHT_MCP_SNAPSHOT_MODE": "optimized"
}
}
}
}
Fine-Tuning:
// Config file approach
{
"snapshotMode": "optimized",
"snapshotOptions": {
"viewportBuffer": 500, // Include elements 500px above/below viewport
"maxTextLength": 50, // Truncate element text to 50 chars
"includeHidden": false, // Exclude hidden elements (recommended)
"minElementSize": { "width": 1, "height": 1 }
}
}
See TOKEN_OPTIMIZATION_GUIDE.md for complete implementation details.
Requirements
- Node.js 18 or newer
- VS Code, Cursor, Windsurf, Claude Desktop, Goose or any other MCP client
Getting started
First, install the Playwright MCP server with your client.
Standard config works in most of the tools:
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": [
"@playwright/mcp@latest"
]
}
}
}
Claude Code
Use the Claude Code CLI to add the Playwright MCP server:
claude mcp add playwright npx @playwright/mcp@latest
Claude Desktop
Follow the MCP install guide, use the standard config above.
Codex
Create or edit the configuration file ~/.codex/config.toml and add:
[mcp_servers.playwright]
command = "npx"
args = ["@playwright/mcp@latest"]
For more information, see the Codex MCP documentation.
Cursor
Click the button to install:
Or install manually:
Go to Cursor Settings -> MCP -> Add new MCP Server. Name to your liking, use command type with the command npx @playwright/mcp. You can also verify config or add command like arguments via clicking Edit.
Gemini CLI
Follow the MCP install guide, use the standard config above.
Goose
Click the button to install:
Or install manually:
Go to Advanced settings -> Extensions -> Add custom extension. Name to your liking, use type STDIO, and set the command to npx @playwright/mcp. Click "Add Extension".
LM Studio
Click the button to install:
Or install manually:
Go to Program in the right sidebar -> Install -> Edit mcp.json. Use the standard config above.
opencode
Follow the MCP Servers documentation. For example in ~/.config/opencode/opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"playwright": {
"type": "local",
"command": [
"npx",
"@playwright/mcp@latest"
],
"enabled": true
}
}
}
Qodo Gen
Open Qodo Gen chat panel in VSCode or IntelliJ → Connect more tools → + Add new MCP → Paste the standard config above.
Click Save.
VS Code
Click the button to install:
Or install manually:
Follow the MCP install guide, use the standard config above. You can also install the Playwright MCP server using the VS Code CLI:
# For VS Code
code --add-mcp '{"name":"playwright","command":"npx","args":["@playwright/mcp@latest"]}'
After installation, the Playwright MCP server will be available for use with your GitHub Copilot agent in VS Code.
Windsurf
Follow Windsurf MCP documentation. Use the standard config above.
Configuration
Playwright MCP server supports following arguments. They can be provided in the JSON configuration above, as a part of the "args" list:
> npx @playwright/mcp@latest --help
--allowed-origins <origins> semicolon-separated list of origins to allow the
browser to request. Default is to allow all.
--blocked-origins <origins> semicolon-separated list of origins to block the
browser from requesting. Blocklist is evaluated
before allowlist. If used without the allowlist,
requests not matching the blocklist are still
allowed.
--block-service-workers block service workers
--browser <browser> browser or chrome channel to use, possible
values: chrome, firefox, webkit, msedge.
--caps <caps> comma-separated list of additional capabilities
to enable, possible values: vision, pdf.
--cdp-endpoint <endpoint> CDP endpoint to connect to.
--config <path> path to the configuration file.
--device <device> device to emulate, for example: "iPhone 15"
--executable-path <path> path to the browser executable.
--extension Connect to a running browser instance
(Edge/Chrome only). Requires the "Playwright MCP
Bridge" browser extension to be installed.
--headless run browser in headless mode, headed by default
--host <host> host to bind server to. Default is localhost. Use
0.0.0.0 to bind to all interfaces.
--ignore-https-errors ignore https errors
--isolated keep the browser profile in memory, do not save
it to disk.
--image-responses <mode> whether to send image responses to the client.
Can be "allow" or "omit", Defaults to "allow".
--no-sandbox disable the sandbox for all process types that
are normally sandboxed.
--output-dir <path> path to the directory for output files.
--port <port> port to listen on for SSE transport.
--proxy-bypass <bypass> comma-separated domains to bypass proxy, for
example ".com,chromium.org,.domain.com"
--proxy-server <proxy> specify proxy server, for example
"http://myproxy:3128" or "socks5://myproxy:8080"
--save-session Whether to save the Playwright MCP session into
the output directory.
--save-trace Whether to save the Playwright Trace of the
session into the output directory.
--storage-state <path> path to the storage state file for isolated
sessions.
--user-agent <ua string> specify user agent string
--user-data-dir <path> path to the user data directory. If not
specified, a temporary directory will be created.
--viewport-size <size> specify browser viewport size in pixels, for
example "1280, 720"
User profile
You can run Playwright MCP with persistent profile like a regular browser (default), in isolated contexts for testing sessions, or connect to your existing browser using the browser extension.
Persistent profile
All the logged in information will be stored in the persistent profile, you can delete it between sessions if you'd like to clear the offline state.
Persistent profile is located at the following locations and you can override it with the --user-data-dir argument.
# Windows
%USERPROFILE%\AppData\Local\ms-playwright\mcp-{channel}-profile
# macOS
- ~/Library/Caches/ms-playwright/mcp-{channel}-profile
# Linux
- ~/.cache/ms-playwright/mcp-{channel}-profile
Isolated
In the isolated mode, each session is started in the isolated profile. Every time you ask MCP to close the browser,
the session is closed and all the storage state for this session is lost. You can provide initial storage state
to the browser via the config's contextOptions or via the --storage-state argument. Learn more about the storage
state here.
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": [
"@playwright/mcp@latest",
"--isolated",
"--storage-state={path/to/storage.json}"
]
}
}
}
Browser Extension
The Playwright MCP Chrome Extension allows you to connect to existing browser tabs and leverage your logged-in sessions and browser state. See extension/README.md for installation and setup instructions.
Configuration file
The Playwright MCP server can be configured using a JSON configuration file. You can specify the configuration file
using the --config command line option:
npx @playwright/mcp@latest --config path/to/config.json
Configuration file schema
{
// Browser configuration
browser?: {
// Browser type to use (chromium, firefox, or webkit)
browserName?: 'chromium' | 'firefox' | 'webkit';
// Keep the browser profile in memory, do not save it to disk.
isolated?: boolean;
// Path to user data directory for browser profile persistence
userDataDir?: string;
// Browser launch options (see Playwright docs)
// @see https://playwright.dev/docs/api/class-browsertype#browser-type-launch
launchOptions?: {
channel?: string; // Browser channel (e.g. 'chrome')
headless?: boolean; // Run in headless mode
executablePath?: string; // Path to browser executable
// ... other Playwright launch options
};
// Browser context options
// @see https://playwright.dev/docs/api/class-browser#browser-new-context
contextOptions?: {
viewport?: { width: number, height: number };
// ... other Playwright context options
};
// CDP endpoint for connecting to existing browser
cdpEndpoint?: string;
// Remote Playwright server endpoint
remoteEndpoint?: string;
},
// Server configuration
server?: {
port?: number; // Port to listen on
host?: string; // Host to bind to (default: localhost)
},
// List of additional capabilities
capabilities?: Array<
'tabs' | // Tab management
'install' | // Browser installation
'pdf' | // PDF generation
'vision' | // Coordinate-based interactions
>;
// Directory for output files
outputDir?: string;
// Network configuration
network?: {
// List of origins to allow the browser to request. Default is to allow all. Origins matching both `allowedOrigins` and `blockedOrigins` will be blocked.
allowedOrigins?: string[];
// List of origins to block the browser to request. Origins matching both `allowedOrigins` and `blockedOrigins` will be blocked.
blockedOrigins?: string[];
};
/**
* Whether to send image responses to the client. Can be "allow" or "omit".
* Defaults to "allow".
*/
imageResponses?: 'allow' | 'omit';
}
Standalone MCP server
When running headed browser on system w/o display or from worker processes of the IDEs,
run the MCP server from environment with the DISPLAY and pass the --port flag to enable HTTP transport.
npx @playwright/mcp@latest --port 8931
And then in MCP client config, set the url to the HTTP endpoint:
{
"mcpServers": {
"playwright": {
"url": "http://localhost:8931/mcp"
}
}
}
Docker
NOTE: The Docker implementation only supports headless chromium at the moment.
{
"mcpServers": {
"playwright": {
"command": "docker",
"args": ["run", "-i", "--rm", "--init", "--pull=always", "mcr.microsoft.com/playwright/mcp"]
}
}
}
You can build the Docker image yourself.
docker build -t mcr.microsoft.com/playwright/mcp .
Programmatic usage
import http from 'http';
import { createConnection } from '@playwright/mcp';
import { SSEServerTransport } from '@modelcontextprotocol/sdk/server/sse.js';
http.createServer(async (req, res) => {
// ...
// Creates a headless Playwright MCP server with SSE transport
const connection = await createConnection({ browser: { launchOptions: { headless: true } } });
const transport = new SSEServerTransport('/messages', res);
await connection.sever.connect(transport);
// ...
});
Tools
Core automation
- browser_click
- Title: Click
- Description: Perform click on a web page
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementref(string): Exact target element reference from the page snapshotdoubleClick(boolean, optional): Whether to perform a double click instead of a single clickbutton(string, optional): Button to click, defaults to left
- Read-only: false
- browser_close
- Title: Close browser
- Description: Close the page
- Parameters: None
- Read-only: true
- browser_connect_roxy
- Title: Connect to RoxyBrowser
- Description: Connect to RoxyBrowser using CDP WebSocket endpoint
- Parameters:
cdpEndpoint(string): The CDP WebSocket endpoint URL from RoxyBrowser (e.g., ws://127.0.0.1:59305/devtools/browser/4d876b0b-6adc-4e9f-b572-bb68ff02a199)
- Read-only: false
- browser_console_messages
- Title: Get console messages
- Description: Returns all console messages
- Parameters: None
- Read-only: true
- browser_drag
- Title: Drag mouse
- Description: Perform drag and drop between two elements
- Parameters:
startElement(string): Human-readable source element description used to obtain the permission to interact with the elementstartRef(string): Exact source element reference from the page snapshotendElement(string): Human-readable target element description used to obtain the permission to interact with the elementendRef(string): Exact target element reference from the page snapshot
- Read-only: false
- browser_evaluate
- Title: Evaluate JavaScript
- Description: Evaluate JavaScript expression on page or element
- Parameters:
function(string): () => { /* code / } or (element) => { / code */ } when element is providedelement(string, optional): Human-readable element description used to obtain permission to interact with the elementref(string, optional): Exact target element reference from the page snapshot
- Read-only: false
- browser_file_upload
- Title: Upload files
- Description: Upload one or multiple files
- Parameters:
paths(array): The absolute paths to the files to upload. Can be a single file or multiple files.
- Read-only: false
- browser_handle_dialog
- Title: Handle a dialog
- Description: Handle a dialog
- Parameters:
accept(boolean): Whether to accept the dialog.promptText(string, optional): The text of the prompt in case of a prompt dialog.
- Read-only: false
- browser_hover
- Title: Hover mouse
- Description: Hover over element on page
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementref(string): Exact target element reference from the page snapshot
- Read-only: true
- browser_navigate
- Title: Navigate to a URL
- Description: Navigate to a URL
- Parameters:
url(string): The URL to navigate to
- Read-only: false
- browser_navigate_back
- Title: Go back
- Description: Go back to the previous page
- Parameters: None
- Read-only: true
- browser_navigate_forward
- Title: Go forward
- Description: Go forward to the next page
- Parameters: None
- Read-only: true
- browser_network_requests
- Title: List network requests
- Description: Returns all network requests since loading the page
- Parameters: None
- Read-only: true
- browser_press_key
- Title: Press a key
- Description: Press a key on the keyboard
- Parameters:
key(string): Name of the key to press or a character to generate, such asArrowLeftora
- Read-only: false
- browser_resize
- Title: Resize browser window
- Description: Resize the browser window
- Parameters:
width(number): Width of the browser windowheight(number): Height of the browser window
- Read-only: true
- browser_select_option
- Title: Select option
- Description: Select an option in a dropdown
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementref(string): Exact target element reference from the page snapshotvalues(array): Array of values to select in the dropdown. This can be a single value or multiple values.
- Read-only: false
- browser_snapshot
- Title: Page snapshot
- Description: Capture accessibility snapshot of the current page, this is better than screenshot
- Parameters: None
- Read-only: true
- browser_take_screenshot
- Title: Take a screenshot
- Description: Take a screenshot of the current page. You can't perform actions based on the screenshot, use browser_snapshot for actions.
- Parameters:
type(string, optional): Image format for the screenshot. Default is png.filename(string, optional): File name to save the screenshot to. Defaults topage-{timestamp}.{png|jpeg}if not specified.element(string, optional): Human-readable element description used to obtain permission to screenshot the element. If not provided, the screenshot will be taken of viewport. If element is provided, ref must be provided too.ref(string, optional): Exact target element reference from the page snapshot. If not provided, the screenshot will be taken of viewport. If ref is provided, element must be provided too.fullPage(boolean, optional): When true, takes a screenshot of the full scrollable page, instead of the currently visible viewport. Cannot be used with element screenshots.
- Read-only: true
- browser_type
- Title: Type text
- Description: Type text into editable element
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementref(string): Exact target element reference from the page snapshottext(string): Text to type into the elementsubmit(boolean, optional): Whether to submit entered text (press Enter after)slowly(boolean, optional): Whether to type one character at a time. Useful for triggering key handlers in the page. By default entire text is filled in at once.
- Read-only: false
- browser_wait_for
- Title: Wait for
- Description: Wait for text to appear or disappear or a specified time to pass
- Parameters:
time(number, optional): The time to wait in secondstext(string, optional): The text to wait fortextGone(string, optional): The text to wait for to disappear
- Read-only: true
Tab management
- browser_tab_close
- Title: Close a tab
- Description: Close a tab
- Parameters:
index(number, optional): The index of the tab to close. Closes current tab if not provided.
- Read-only: false
- browser_tab_list
- Title: List tabs
- Description: List browser tabs
- Parameters: None
- Read-only: true
- browser_tab_new
- Title: Open a new tab
- Description: Open a new tab
- Parameters:
url(string, optional): The URL to navigate to in the new tab. If not provided, the new tab will be blank.
- Read-only: true
- browser_tab_select
- Title: Select a tab
- Description: Select a tab by index
- Parameters:
index(number): The index of the tab to select
- Read-only: true
Browser installation
- browser_install
- Title: Install the browser specified in the config
- Description: Install the browser specified in the config. Call this if you get an error about the browser not being installed.
- Parameters: None
- Read-only: false
Coordinate-based (opt-in via --caps=vision)
- browser_mouse_click_xy
- Title: Click
- Description: Click left mouse button at a given position
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementx(number): X coordinatey(number): Y coordinate
- Read-only: false
- browser_mouse_drag_xy
- Title: Drag mouse
- Description: Drag left mouse button to a given position
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementstartX(number): Start X coordinatestartY(number): Start Y coordinateendX(number): End X coordinateendY(number): End Y coordinate
- Read-only: false
- browser_mouse_move_xy
- Title: Move mouse
- Description: Move mouse to a given position
- Parameters:
element(string): Human-readable element description used to obtain permission to interact with the elementx(number): X coordinatey(number): Y coordinate
- Read-only: true
PDF generation (opt-in via --caps=pdf)
- browser_pdf_save
- Title: Save as PDF
- Description: Save page as PDF
- Parameters:
filename(string, optional): File name to save the pdf to. Defaults topage-{timestamp}.pdfif not specified.
- Read-only: true