Playwright MCP: let your AI assistant use a real browser

• Opens a web page in the browser at the address you give

  Loads the URL you give into the browser so the assistant can work on that page.

  Try: “Open https://example.com/signup in the browser”

  Network · browser_navigate

• Reads the text, links and buttons on the current page so the assistant can see it

  Captures the structure of the current page, which the assistant uses to decide what to click or fill next.

  Try: “What is on the page right now?”

  Files · browser_snapshot

• Takes a picture of the current page or one element on it

  Saves an image of the visible page, the full scrollable page, or a single element, as PNG, JPEG or WebP.

  Try: “Take a full-page screenshot of the pricing page”

  Files · browser_take_screenshot

• Clicks a button, link or other element on the page

  Clicks one element, with an option for double-click and for the left, right or middle button.

  Try: “Click the Sign up button and tell me what happens”

  Write · browser_click

• Types text into a field, optionally pressing Enter afterward

  Enters text into an editable field, either all at once or one character at a time for pages that react to each key.

  Try: “Type my search term into the search box and press Enter”

  Write · browser_type

• Fills in several form fields in one step

  Fills a list of form fields together, so a multi-field form does not need one action per field.

  Try: “Fill in the contact form with the name and email from my notes”

  Write · browser_fill_form

• Picks one or more options in a dropdown menu

  Chooses one or several values in a dropdown on the page.

  Try: “Set the country dropdown on the checkout form to Canada”

  Write · browser_select_option

• Presses a keyboard key such as Enter, Tab or an arrow key

  Sends one key press to the page, or a single character.

  Try: “Press Tab to move to the next field”

  Write · browser_press_key

• Moves the mouse over an element to open a menu or tooltip

  Puts the mouse pointer over an element without clicking, which reveals hover menus and tooltips.

  Try: “Hover over the Products menu and list the sub-items”

  Write · browser_hover

• Drags one element and drops it onto another

  Performs a drag from a start element to an end element on the page.

  Try: “Drag the Done card into the Shipped column”

  Write · browser_drag

• Drops files or data onto an element, as if dragged in from outside the page

  Simulates dropping files from your computer, or MIME-typed text and links, onto a page element.

  Try: “Drop report.pdf from my project folder onto the upload area”

  Write · browser_drop

• Uploads one or more files from your computer into a page

  Attaches the files at the paths you give to the page's file picker. If no paths are given, the picker is cancelled.

  Try: “Upload invoice-march.pdf to the expense form”

  Write · browser_file_upload

• Accepts or dismisses a pop-up alert or confirm box

  Answers a browser alert, confirm or prompt box, with optional text for a prompt.

  Try: “Accept the confirmation box that asks to delete the draft”

  Write · browser_handle_dialog

• Goes back to the previous page in the browser history

  Moves the current tab one step back through its history.

  Try: “Go back to the search results”

  Network · browser_navigate_back

• Waits for text to appear or disappear, or for a set number of seconds

  Pauses the session until the text you name shows up or goes away, or for up to 30 seconds.

  Try: “Wait until the message Order confirmed appears”

  Read · browser_wait_for

• Changes the browser window to a width and height you choose

  Sets the browser window size in pixels, useful for checking a mobile layout.

  Try: “Resize the window to 390 by 844 to check the mobile layout”

  Write · browser_resize

• Switches the page between light and dark mode, reduced motion or print styles

  Overrides color scheme, motion, contrast, forced colors and the CSS media type. Leave a setting out to keep it as it is.

  Try: “Show the homepage in dark mode with reduced motion on”

  Write · browser_emulate_media

• Lists, opens, closes or switches between browser tabs

  Manages the open tabs: list them, open a new one at a URL, close one, or select one by its position.

  Try: “Open a new tab with my staging site and switch to it”

  Write · browser_tabs

• Closes the current browser page

  Closes the page the assistant is working on.

  Try: “Close the browser when you are done”

  Write · browser_close

• Returns the messages the page wrote to its developer console

  Lists console output such as logs, warnings and errors, filtered by level, and can save them to a file.

  Try: “Show any errors the checkout page logged to the console”

  Files · browser_console_messages

• Lists the network requests the page made, with an optional URL filter

  Returns a numbered list of requests since the page loaded, so you can see which calls a page made and where they went.

  Try: “List the API calls the dashboard made that match /api/orders”

  Files · browser_network_requests

• Shows the headers and body of one request from the network list

  Returns the full details of a single request by its number from the list, or only one part of it.

  Try: “Show the response body of request 12”

  Files · browser_network_request

• Runs a JavaScript expression on the page or on one element and returns the result

  Executes a short JavaScript function inside the page and returns its value, optionally saving it to a file.

  Try: “Run a script that counts the images on this page”

  Runs code · browser_evaluate

• Runs a custom automation script with full control of the page. Can run any code

  Executes a Playwright-style script inside the automation server, from text or a file. It can run arbitrary JavaScript there, so use it only with code you trust.

  Try: “Run this script that clicks the first result and returns its title”

  Runs code · browser_run_code_unsafe

• Takes the browser offline or back online. Offline makes every request fail

  Sets the network state to offline to test how a page behaves without connectivity, or back to online.

  Try: “Take the browser offline and reload the app”

  Write · browser_network_state_set

• Fakes responses for network requests that match a URL pattern

  Intercepts requests matching a pattern and returns a status, body, content type or headers you set instead of the real response.

  Try: “Make every call to /api/users return an empty list”

  Write · browser_route

• Lists the fake network responses that are active right now

  Shows each active interception rule set with the route tool.

  Try: “Which fake responses are still active?”

  Read · browser_route_list

• Removes fake network responses for one pattern, or all of them

  Deletes an interception rule by its URL pattern, or every rule when no pattern is given.

  Try: “Remove the fake response for /api/users”

  Write · browser_unroute

• Lists the cookies stored in the browser, optionally by domain or path

  Returns the cookies the browser holds, filtered by domain or path if you give one.

  Try: “List the cookies the site set for example.com”

  Read · browser_cookie_list

• Gets one cookie by its name

  Returns the value and details of the named cookie.

  Try: “What is the value of the session cookie?”

  Read · browser_cookie_get

• Sets a cookie with an optional domain, path, expiry and security flags

  Creates or replaces a cookie with the name and value you give, plus any flags you set.

  Try: “Set a cookie named beta to true on example.com”

  Write · browser_cookie_set

• Deletes one cookie by name, optionally only on one domain or path

  Removes the named cookie, limited to a domain or path if you give one.

  Try: “Delete the tracking cookie from example.com”

  Delete · browser_cookie_delete

• Deletes every cookie in the browser

  Removes all cookies, so the next page load starts with no saved sessions.

  Try: “Clear all cookies so I start logged out”

  Delete · browser_cookie_clear

• Lists every item in the page's local storage

  Returns all key and value pairs in local storage for the current page.

  Try: “Show everything the app keeps in local storage”

  Read · browser_localstorage_list

• Reads one local storage item by its key

  Returns the value stored under the key you name.

  Try: “What is saved under the theme key in local storage?”

  Read · browser_localstorage_get

• Saves a value under a key in local storage

  Writes the value you give under the key you give in local storage.

  Try: “Set the theme key in local storage to dark”

  Write · browser_localstorage_set

• Deletes one local storage item by its key

  Removes the value stored under the named key.

  Try: “Remove the draft key from local storage”

  Delete · browser_localstorage_delete

• Deletes every local storage item for the page

  Removes all local storage entries for the current page.

  Try: “Clear local storage for this site”

  Delete · browser_localstorage_clear

• Lists every item in the page's session storage

  Returns all key and value pairs in session storage for the current tab.

  Try: “Show what this tab has in session storage”

  Read · browser_sessionstorage_list

• Reads one session storage item by its key

  Returns the value stored under the key you name in session storage.

  Try: “What is the cart id in session storage?”

  Read · browser_sessionstorage_get

• Saves a value under a key in session storage

  Writes the value you give under the key you give in session storage.

  Try: “Set step to 2 in session storage”

  Write · browser_sessionstorage_set

• Deletes one session storage item by its key

  Removes the value stored under the named key in session storage.

  Try: “Remove the step key from session storage”

  Delete · browser_sessionstorage_delete

• Deletes every session storage item for the page

  Removes all session storage entries for the current tab.

  Try: “Clear session storage in this tab”

  Delete · browser_sessionstorage_clear

• Saves cookies and local storage to a file you can reuse later

  Writes the current cookies and local storage to a file, so a later session can start logged in.

  Try: “Save my logged-in browser state to login-state.json”

  Files · browser_storage_state

• Restores cookies and local storage from a saved file, clearing what is there

  Reads a saved state file and loads its cookies and local storage. Existing cookies and local storage are cleared first.

  Try: “Restore the login state from login-state.json”

  Delete · browser_set_storage_state

• Records the actions you take in the browser and turns them into code

  Starts capturing the clicks and typing you do in the browser so they can be replayed as code.

  Try: “Start recording while I walk through the checkout flow”

  Read · browser_start_recording

• Stops the recording and returns the captured actions as code

  Ends the recording started earlier and returns the recorded steps as code.

  Try: “Stop recording and give me the code”

  Read · browser_stop_recording

• Starts recording a trace of the browser session

  Begins a trace that captures what the browser does, for debugging later.

  Try: “Start a trace before I reproduce the bug”

  Read · browser_start_tracing

• Stops the trace recording

  Ends the trace started earlier.

  Try: “Stop the trace now”

  Read · browser_stop_tracing

• Starts recording a video of the browser session

  Begins a screen recording of the page, saved to a file with an optional frame rate, size and animated cursor.

  Try: “Start recording a video of the signup flow”

  Files · browser_start_video

• Stops the video recording

  Ends the video started earlier.

  Try: “Stop the video recording”

  Read · browser_stop_video

• Adds a titled chapter card to the video recording

  Shows a full-screen card with a title and optional description in the video, for splitting a recording into parts.

  Try: “Add a chapter called Step 2: Payment to the video”

  Write · browser_video_chapter

• Labels each action on screen with a callout while the video records

  Turns on callouts that name each action and can animate a cursor between action points.

  Try: “Show a label for every click in the video”

  Write · browser_video_show_actions

• Turns off the action callouts on screen

  Stops annotating actions on the page.

  Try: “Hide the action labels”

  Write · browser_video_hide_actions

• Opens annotation mode so you can draw on the page, then returns the marked-up image

  Opens the page in annotation mode and waits for you to draw. Returns the annotated screenshot and a list of your marks.

  Try: “Let me mark up the broken layout, then send it back to you”

  Read · browser_annotate

• Draws a lasting outline around an element on the page

  Adds a highlight box to one element, with optional custom CSS, until it is removed.

  Try: “Outline the price field in red so I can find it”

  Write · browser_highlight

• Removes the highlight drawn around an element

  Takes off a highlight added earlier for the same element.

  Try: “Remove the red outline from the price field”

  Write · browser_hide_highlight

• Resumes a paused script, optionally pausing again before each step

  Continues script execution after a pause. Set step to pause before every next action, for slow debugging.

  Try: “Resume the script and pause before the next click”

  Runs code · browser_resume

• Saves the current page as a PDF file

  Writes the current page to a PDF, at the file name you give or an automatic name.

  Try: “Save this invoice page as a PDF”

  Files · browser_pdf_save

• Clicks at exact screen coordinates on the page

  Clicks at an X and Y position, for elements that have no usable reference. Supports double-click and button choice.

  Try: “Click at position 420, 310 on the canvas”

  Write · browser_mouse_click_xy

• Moves the mouse pointer to exact screen coordinates

  Moves the pointer to an X and Y position without clicking.

  Try: “Move the mouse to 100, 200”

  Write · browser_mouse_move_xy

• Presses a mouse button down at the current pointer position

  Holds the chosen mouse button down. Pair with the release tool to make a manual press.

  Try: “Press the mouse button down here”

  Write · browser_mouse_down

• Releases a mouse button at the current pointer position

  Lets go of the chosen mouse button.

  Try: “Release the mouse button”

  Write · browser_mouse_up

• Drags from one set of screen coordinates to another

  Holds the left button at a start X and Y and drags to an end X and Y.

  Try: “Drag the slider from 50, 300 to 250, 300”

  Write · browser_mouse_drag_xy

• Scrolls the page with the mouse wheel by a set amount

  Scrolls horizontally or vertically by the number of pixels you give.

  Try: “Scroll the page down by 800 pixels”

  Write · browser_mouse_wheel

• Creates a test locator for an element so it can be used in test code

  Produces the code expression that finds a given element, for writing automated tests.

  Try: “Give me a test locator for the Submit button”

  Read · browser_generate_locator

• Checks that an element with a given role and name is visible on the page

  Asserts that an element with the role and accessible name you give is visible.

  Try: “Check that the Save button is visible”

  Read · browser_verify_element_visible

• Checks that a list on the page shows the items you expect

  Asserts that the named list contains the listed items.

  Try: “Check that the cart list shows Shirt and Socks”

  Read · browser_verify_list_visible

• Checks that a piece of text is visible on the page

  Asserts that the text you give is visible. Prefer the element check when you can.

  Try: “Check that the text Payment received is on the page”

  Read · browser_verify_text_visible

• Checks the current value of a form field or checkbox

  Asserts that a field holds the value you give. For a checkbox, use true or false.

  Try: “Check that the newsletter box is ticked”

  Read · browser_verify_value

• Used by the assistant behind the scenes

  Shows the final settings the browser tool is running with

  Returns the merged result of command options, environment variables and the config file.

  Try: “Which browser capabilities are turned on right now?”

  Read · browser_get_config

• Searches the page for text or a pattern to find the element to act on

  Searches the page structure for text or a regular expression and returns matches with a few lines of context. The assistant uses it to locate an element before acting on it.

  Try: “Find the element that says Add to cart”

  Files · browser_find

Watch out for

• It can run any code on your computer

  One tool runs JavaScript in the server process, and the README calls it equivalent to remote code execution. Keep approval switched on for it. github.com

• Logins are saved on disk by default

  The default profile keeps cookies and sessions between runs. Use the isolated option for throwaway sessions, or delete the profile folder. github.com

• Extension mode drives your real browser

  With the extension option the assistant works in your open Chrome or Edge tabs, signed in as you. Only use it when you are watching. github.com

• Microsoft says it is not a security boundary

  Origin filters and the file-access limit are guardrails that can be worked around. Do not rely on them to contain a risky site. github.com

• It can click, submit and change site data

  The core tools click, type and submit forms on real sites, and opt-in tools can set or clear cookies. Confirm actions on accounts you care about. github.com

• Keep HTTP mode on localhost

  The port option starts a local web server, and binding to 0.0.0.0 exposes it to your network. Update past 0.0.40 for DNS rebinding protection. github.com

Security advisories

In their words