browser_click
Perform click on a web page
LABSCO SUMMARY
An agent can use a website the way a person does.
What it takes off you is the driving: open the page, find the control, fill the form, then check what actually happened — against the live page rather than fetched HTML. Every action returns the Playwright line that performed it, so the session reads back as code afterwards.
What it is
Microsoft's own MCP server for Playwright. It drives a real browser and works off the page's accessibility snapshot, so an agent addresses named elements instead of coordinates. Twenty-four tools are on by default; another forty-five sit behind --caps.
What you get
- Navigate, click, hover, type, press keys, select options, drag and drop, upload a file, handle a dialog, and fill a whole form in one call (
browser_navigate,browser_click,browser_fill_form,browser_file_upload,browser_handle_dialog) - An accessibility snapshot of the page, plus a search over it that returns matching nodes with their surrounding context (
browser_snapshot,browser_find) - Console messages, a numbered list of network requests, and the full detail of any one of them (
browser_console_messages,browser_network_requests,browser_network_request) - Screenshots, resizing, waiting for text to appear or disappear, tab management, and JavaScript evaluated on the page or a single element (
browser_take_screenshot,browser_wait_for,browser_tabs,browser_evaluate) - Seventeen storage tools behind
--caps=storage: cookies and local and session storage read, set, deleted and cleared, plus a whole storage state saved and restored — which is how a logged-in session is reused instead of re-entered - Four network tools behind
--caps=network: routes that mock a URL pattern with your own status, body and headers, listed and removed, and the browser switched offline - Eleven DevTools tools behind
--caps=devtools: tracing started and stopped, video recording with chapters, and elements highlighted or annotated on the page - Five assertion tools behind
--caps=testingfor checking an element, a list, a text or a value is visible, plus a locator generated for it — the pieces of a test rather than a click - Coordinate-based mouse control behind
--caps=vision, PDF export behind--caps=pdf, and the resolved configuration read back behind--caps=config
Requirements
Node.js 18 or newer and npx — nothing else. It starts as a local process and talks to the client over stdio, so the browser it drives is the one on that machine. One thing to know before running two clients at once: a persistent profile can only be used by one browser instance at a time, so give each extra client --isolated or its own --user-data-dir. Extra capabilities come from --caps as a comma-separated list, or the PLAYWRIGHT_MCP_CAPS environment variable.
Setup effort
One command — npx @playwright/mcp@latest
USE CASES
01
Checking a flow still works
Walk the signup, the checkout or the settings page and read the snapshot at each step. Because every action comes back as the Playwright line that performed it, the session you just ran is most of the test you were going to write.
02
Reproducing a reported bug
Follow the steps in the report, then read the console messages and the network requests from the same session. The evidence and the reproduction arrive together instead of being reconstructed afterwards.
03
Getting through a long form
Fill every field in one call from the snapshot's named elements rather than one type-and-click at a time. Fewer round trips, and no coordinates to go stale when the layout shifts.
04
Testing the page when the API is down or slow
Turn on the network capability and mock the endpoint with your own status and body, or flip the browser offline. That gets you the error state without breaking a real backend to see it.
05
Not logging in again every session
With the storage capability on, save the storage state once after signing in and restore it at the start of later runs. Cookies and both web storages come back with it.
06
Pulling data off a page with no API
Snapshot the page, search it for the pattern you want, and evaluate JavaScript against the matching element when the structure needs unpicking. It reads the rendered page, so client-side rendering is not an obstacle.
07
Turning a session into a test
With the testing capability on, assert that the element, list, text or value is what it should be, and generate a locator for it. What you get back is test code rather than a transcript.
24 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
browser_close
Close the page
browser_console_messages
Returns all console messages
browser_drag
Perform drag and drop between two elements
browser_drop
Drop files or MIME-typed data onto an element, as if dragged from outside the page. At least one of "paths" or "data" must be provided.
browser_evaluate
Evaluate JavaScript expression on page or element
browser_file_upload
Upload one or multiple files
browser_fill_form
Fill multiple form fields
browser_find
Search the accessibility snapshot of the current page for text or a regular expression. Returns matching snapshot nodes with a few lines of surrounding context (like search snippets), each shown under its path from the root of the tree, which is cheaper than capturing the whole snapshot when you only need to locate an element and its ref.
browser_handle_dialog
Handle a dialog
browser_hover
Hover over element on page
browser_navigate
Navigate to a URL
browser_navigate_back
Go back to the previous page in the history
browser_network_request
Returns full details (headers and body) of a single network request, or a single part if part is set. Use the number from browser_network_requests.
browser_network_requests
Returns a numbered list of network requests since loading the page. Use browser_network_request with the number to get full details.
browser_press_key
Press a key on the keyboard
browser_resize
Resize the browser window
browser_run_code_unsafe
Run a Playwright code snippet. Unsafe: executes arbitrary JavaScript in the Playwright server process and is RCE-equivalent.
browser_select_option
Select an option in a dropdown
browser_snapshot
Capture accessibility snapshot of the current page, this is better than screenshot
browser_tabs
List, create, close, or select a browser tab.
browser_take_screenshot
Take a screenshot of the current page. You can't perform actions based on the screenshot, use browser_snapshot for actions.
browser_type
Type text into editable element
browser_wait_for
Wait for text to appear or disappear or a specified time to pass
BEFORE YOU INSTALL
What to know first
Tools you get
24 tools when we connected; the project documents 69. Only the Core automation and Tab management groups load by default. The rest are opt-in with --caps=config, network, storage, devtools, vision, pdf or testing, and --read-only narrows the set further.
Requires
- npx on PATH
- Node.js 18 or later
Runs code
One of the tools runs code rather than driving the browser: browser_run_code_unsafe takes a Playwright snippet and, in its own description, "executes arbitrary JavaScript in the Playwright server process and is RCE-equivalent." Page-level JavaScript has its own tool — browser_evaluate — which runs on the page or on a single element.
INSTALLATION
Connect your client
Server setup
TERMINAL
npx @playwright/mcp@latestMCP Clients
Amp
AMP · JSON
{
"amp.mcpServers": {
"playwright": {
"command": "npx",
"args": [
"@playwright/mcp@latest"
]
}
}
}VS Code
VS CODE · URL
https://insiders.vscode.dev/redirect?url=vscode%3Amcp%2Finstall%3F%257B%2522name%2522%253A%2522playwright%2522%252C%2522command%2522%253A%2522npx%2522%252C%2522args%2522%253A%255B%2522%2540playwright%252Fmcp%2540latest%2522%255D%257D
