Release history
Playwright MCP server
46 shown
Fixed serialization of shared browser launches and forwarding of CDP commands in isolated and extension modes.
browser_network_requests — Now returns a numbered list of requests; use the new browser_network_request tool with the index to fetch full details (headers/body) or a single part. Both tools accept a filename option to write the output to a file (#40447, #40454)browser_run_code → browser_run_code_unsafe — Renamed to make the sandbox-escape implication explicit (#40496)browser_drop — Expose Locator.drop as an MCP tool for completing drag-and-drop operations onto a target (#40367)browser_network_requests — Added responseBody and responseHeaders options; binary bodies are rendered as a placeholder using mime-type detection (microsoft/playwright@f0be0b17)browser_evaluate — Accept plain expressions in addition to function bodies (#39979)Minor fixes and improvements.
browser_network_state_set — Toggle network offline mode for testing connectivity scenarios (#39459)browser_video_chapter — Add a chapter marker to the video recording, showing a full-screen chapter card with blurred backdrop (#39891)browser_mouse_click_xy — Added button, clickCount, and delay options for more precise mouse interaction control (#39368, #39465)browser_network_requests — Added filtering support and optional headers/body fields in responses (#39672)mcpName field to package for MCP Registry ownership verification (#1432)--snapshot-mode=none is set (#39378)toWellFormed() compatibility with Node.js 18 (#39674)cdpHeaders from environment overriding config file values (#39866)browser_set_storage_state tool (#39881)Minor fixes and improvements.
--no-sandbox → --no-chromium-sandbox rename.--no-sandbox.--[no-]sandbox to --[no-]chromium-sandbox for clarity and to avoid ambiguity with other browsers.Browser profiles are now in-memory by default — every session starts clean with no leftover state.
Use --persistent to opt into a persistent profile, or --profile=<path> to specify a custom profile directory:
> playwright-cli open example.com # incognito, clean slate
> playwright-cli open example.com --persistent # persistent profile
> playwright-cli open example.com --profile=./my-data # custom profile directory
The --session flag has been replaced with the shorter -s=. Session lifecycle is simplified — there is no longer a "stopped" state; sessions are either running or gone.
> playwright-cli -s=myapp open example.com
> playwright-cli -s=myapp click e5
> playwright-cli -s=myapp close
> playwright-cli -s=myapp delete-data
New session management commands replace the old session-* family:
> playwright-cli list # list all sessions
> playwright-cli close-all # close all browsers
> playwright-cli kill-all # forcefully kill all browser processes
Each workspace now gets its own daemon process, preventing cross-project interference and enabling direct daemon startup for faster, more reliable operation.
Full introspection and manipulation of browser storage, enabling reproducible auth flows, debugging, and stateful automation.
> playwright-cli state-save auth.json
> playwright-cli state-load auth.json
Inspect and manage cookies:
> playwright-cli cookie-list
> playwright-cli cookie-get session_id
> playwright-cli cookie-set session_id abc123 --domain=example.com
> playwright-cli cookie-delete session_id
> playwright-cli cookie-clear
Work directly with Web Storage APIs:
# LocalStorage
> playwright-cli localstorage-list
> playwright-cli localstorage-get theme
> playwright-cli localstorage-set theme dark
> playwright-cli localstorage-clear
# SessionStorage
> playwright-cli sessionstorage-list
> playwright-cli sessionstorage-set wizardStep 3
> playwright-cli sessionstorage-clear
Perfect for:
Intercept, mock, modify, or block network requests directly from the CLI — no test harness required.
Mock responses by URL pattern with custom status codes, bodies, headers, or request mutations.
# Mock with custom status
> playwright-cli route "**/*.jpg" --status=404
# Mock with JSON body
> playwright-cli route "**/api/users" \
--body='[{"id":1,"name":"Alice"}]' \
--content-type=application/json
# Mock with custom headers
> playwright-cli route "**/api/data" \
--body='{"ok":true}' \
--header="X-Custom: value"
# Strip sensitive headers from outgoing requests
> playwright-cli route "**/*" --remove-header=cookie,authorization
Manage active routes:
> playwright-cli route-list
> playwright-cli unroute "**/*.jpg"
> playwright-cli unroute
Use cases:
Routes support glob-style patterns for precise targeting:
**/api/users Exact path match
**/api/*/details Wildcards within paths
**/*.{png,jpg,jpeg} File extension matching
**/search?q=* Query parameter matching
run-codeFor conditional logic, request inspection, response mutation, or timing control, drop down to raw Playwright routing via run-code.
> playwright-cli run-code "async page => {
await page.route('**/api/login', route => {
const body = route.request().postDataJSON();
if (body.username === 'admin') {
route.fulfill({ body: JSON.stringify({ token: 'mock-token' }) });
} else {
route.fulfill({ status: 401, body: JSON.stringify({ error: 'Invalid' }) });
}
});
}"
> playwright-cli run-code "async page => {
await page.route('**/api/user', async route => {
const response = await route.fetch();
const json = await response.json();
json.isPremium = true;
await route.fulfill({ response, json });
});
}"
> playwright-cli run-code "async page => {
await page.route('**/api/offline', route =>
route.abort('internetdisconnected')
);
}"
> playwright-cli install-skills
> playwright-cli install-browser
> playwright-cli config --in-memory
Renamed from --isolated to better reflect behavior:
Infrastructure for Playwright CLI is in, so we are ready to cook the features and skills!
> playwright-cli open example.com
> time playwright-cli snapshot
0.03s user 0.01s system cpu 0.055 total
> playwright-cli video-start
> playwright-cli open playwright.dev
> playwright-cli video-stop
### Result
- [Video](.playwright-cli/video-2026-01-31T00-49-47-976Z.webm)
Also available with the --caps=devtools flag in Playwright MCP.
> playwright-cli config --headed --config=config.json
> playwright-cli open playwright.dev
Countless configuration options
> playwright-cli install --browser=firefox
> playwright-cli config --isolated --browser=firefox --headed
> playwright-cli open playwright.dev
> playwright-cli --session=real-browser snapshot --extension
Share the tab and have fun!
Minor fixes and improvements.
Minor fixes and improvements.
playwright-cli.jsonWe are adding a new token-efficient CLI mode of operation to Playwright with the skills located at playwright-cli. This brings the long-awaited official SKILL-focused CLI mode to our story and makes it more coding agent-friendly.
It is the first snapshot with the essential command set (which is already larger than the original MCP!), but we expect it to grow rapidly. Unlike the token use, that one we expect to go down since snapshots are no longer forced into the LLM!
npm install -g @playwright/mcp@latest
playwright-cli --help
Your agent will be running those, but it does not mean you can't play with it:
playwright-cli open https://demo.playwright.dev/todomvc/ --headed
playwright-cli type "Buy groceries"
playwright-cli press Enter
playwright-cli type "Water flowers"
playwright-cli press Enter
playwright-cli check e21
playwright-cli check e35
playwright-cli press screenshot
Point your agent at the CLI and let it cook. It'll read the skill off playwright-cli --help on its own:
Test the "add todo" flow on https://demo.playwright.dev/todomvc using playwright-cli.
Check playwright-cli --help for available commands.
Claude Code, GitHub copilot and others will let you install the Playwright skills into the agentic loop.
/plugin marketplace add microsoft/playwright-cli
/plugin install playwright-cli
mkdir -p .claude/skills/playwright-cli
curl -o .claude/skills/playwright-cli/SKILL.md \
https://raw.githubusercontent.com/microsoft/playwright-cli/main/skills/playwright-cli/SKILL.md
Playwright CLI is headless by default. If you'd like to see the browser, pass --headed to open:
playwright-cli open https://playwright.dev --headed
Playwright CLI will use a dedicated persistent profile by default. It means that
your cookies and other storage state will be preserved between the calls. You can use different
instances of the browser for different projects with sessions.
Following will result in two browsers with separate profiles being available. Pass --session to
the invocation to talk to a specific browser.
playwright-cli open https://playwright.dev
playwright-cli --session=example open https://example.com
playwright-cli session-list
You can run your coding agent with the PLAYWRIGHT_CLI_SESSION environment variable:
PLAYWRIGHT_CLI_SESSION=todo-app claude .
Or instruct it to prepend --session to the calls.
Manage your sessions as follows:
playwright-cli session-list # list all sessions
playwright-cli session-stop [name] # stop session
playwright-cli session-stop-all # stop all sessions
playwright-cli session-delete [name] # delete session data along with the profiles
playwright-cli open <url> # open url
playwright-cli close # close the page
playwright-cli type <text> # type text into editable element
playwright-cli click <ref> [button] # perform click on a web page
playwright-cli dblclick <ref> [button] # perform double click on a web page
playwright-cli fill <ref> <text> # fill text into editable element
playwright-cli drag <startRef> <endRef> # perform drag and drop between two elements
playwright-cli hover <ref> # hover over element on page
playwright-cli select <ref> <val> # select an option in a dropdown
playwright-cli upload <file> # upload one or multiple files
playwright-cli check <ref> # check a checkbox or radio button
playwright-cli uncheck <ref> # uncheck a checkbox or radio button
playwright-cli snapshot # capture page snapshot to obtain element ref
playwright-cli eval <func> [ref] # evaluate javascript expression on page or element
playwright-cli dialog-accept [prompt] # accept a dialog
playwright-cli dialog-dismiss # dismiss a dialog
playwright-cli resize <w> <h> # resize the browser window
playwright-cli go-back # go back to the previous page
playwright-cli go-forward # go forward to the next page
playwright-cli reload # reload the current page
playwright-cli press <key> # press a key on the keyboard, `a`, `arrowleft`
playwright-cli keydown <key> # press a key down on the keyboard
playwright-cli keyup <key> # press a key up on the keyboard
playwright-cli mousemove <x> <y> # move mouse to a given position
playwright-cli mousedown [button] # press mouse down
playwright-cli mouseup [button] # press mouse up
playwright-cli mousewheel <dx> <dy> # scroll mouse wheel
playwright-cli screenshot [ref] # screenshot of the current page or element
playwright-cli pdf # save page as pdf
playwright-cli tab-list # list all tabs
playwright-cli tab-new [url] # create a new tab
playwright-cli tab-close [index] # close a browser tab
playwright-cli tab-select <index> # select a browser tab
playwright-cli console [min-level] # list console messages
playwright-cli network # list all network requests since loading the page
playwright-cli run-code <code> # run playwright code snippet
playwright-cli tracing-start # start trace recording
playwright-cli tracing-stop # stop trace recording
We are adding a new token-efficient CLI mode of operation to Playwright with the skills located at playwright-cli. This brings the long-awaited official SKILL-focused CLI mode to our story and makes it more coding agent-friendly.
It is the first snapshot with the essential command set (which is already larger than the original MCP!), but we expect it to grow rapidly. Unlike the token use, that one we expect to go down since snapshots are no longer forced into the LLM!
npm install -g @playwright/mcp@latest
playwright-cli --help
Your agent will be running those, but it does not mean you can't play with it:
playwright-cli open https://demo.playwright.dev/todomvc/ --headed
playwright-cli type "Buy groceries"
playwright-cli press Enter
playwright-cli type "Water flowers"
playwright-cli press Enter
playwright-cli check e21
playwright-cli check e35
playwright-cli press screenshot
Point your agent at the CLI and let it cook. It'll read the skill off playwright-cli --help on its own:
Test the "add todo" flow on https://demo.playwright.dev/todomvc using playwright-cli.
Check playwright-cli --help for available commands.
Claude Code, GitHub copilot and others will let you install the Playwright skills into the agentic loop.
/plugin marketplace add microsoft/playwright-cli
/plugin install playwright-cli
mkdir -p .claude/skills/playwright-cli
curl -o .claude/skills/playwright-cli/SKILL.md \
https://raw.githubusercontent.com/microsoft/playwright-cli/main/skills/playwright-cli/SKILL.md
Playwright CLI is headless by default. If you'd like to see the browser, pass --headed to open:
playwright-cli open https://playwright.dev --headed
Playwright CLI will use a dedicated persistent profile by default. It means that
your cookies and other storage state will be preserved between the calls. You can use different
instances of the browser for different projects with sessions.
Following will result in two browsers with separate profiles being available. Pass --session to
the invocation to talk to a specific browser.
playwright-cli open https://playwright.dev
playwright-cli --session=example open https://example.com
playwright-cli session-list
You can run your coding agent with the PLAYWRIGHT_CLI_SESSION environment variable:
PLAYWRIGHT_CLI_SESSION=todo-app claude .
Or instruct it to prepend --session to the calls.
Manage your sessions as follows:
playwright-cli session-list # list all sessions
playwright-cli session-stop [name] # stop session
playwright-cli session-stop-all # stop all sessions
playwright-cli session-delete [name] # delete session data along with the profiles
playwright-cli open <url> # open url
playwright-cli close # close the page
playwright-cli type <text> # type text into editable element
playwright-cli click <ref> [button] # perform click on a web page
playwright-cli dblclick <ref> [button] # perform double click on a web page
playwright-cli fill <ref> <text> # fill text into editable element
playwright-cli drag <startRef> <endRef> # perform drag and drop between two elements
playwright-cli hover <ref> # hover over element on page
playwright-cli select <ref> <val> # select an option in a dropdown
playwright-cli upload <file> # upload one or multiple files
playwright-cli check <ref> # check a checkbox or radio button
playwright-cli uncheck <ref> # uncheck a checkbox or radio button
playwright-cli snapshot # capture page snapshot to obtain element ref
playwright-cli eval <func> [ref] # evaluate javascript expression on page or element
playwright-cli dialog-accept [prompt] # accept a dialog
playwright-cli dialog-dismiss # dismiss a dialog
playwright-cli resize <w> <h> # resize the browser window
playwright-cli go-back # go back to the previous page
playwright-cli go-forward # go forward to the next page
playwright-cli reload # reload the current page
playwright-cli press <key> # press a key on the keyboard, `a`, `arrowleft`
playwright-cli keydown <key> # press a key down on the keyboard
playwright-cli keyup <key> # press a key up on the keyboard
playwright-cli mousemove <x> <y> # move mouse to a given position
playwright-cli mousedown [button] # press mouse down
playwright-cli mouseup [button] # press mouse up
playwright-cli mousewheel <dx> <dy> # scroll mouse wheel
playwright-cli screenshot [ref] # screenshot of the current page or element
playwright-cli pdf # save page as pdf
playwright-cli tab-list # list all tabs
playwright-cli tab-new [url] # create a new tab
playwright-cli tab-close [index] # close a browser tab
playwright-cli tab-select <index> # select a browser tab
playwright-cli console [min-level] # list console messages
playwright-cli network # list all network requests since loading the page
playwright-cli run-code <code> # run playwright code snippet
playwright-cli tracing-start # start trace recording
playwright-cli tracing-stop # stop trace recording
Routine maintenance release for microsoft/playwright-mcp.
cdpTimeout configuration option for CDP connections, allowing control over connection timeouts (default: 30s, 0 disables timeout).playwright install-deps.http:, https:, about:, and data: URLs; file:// is blocked.setInputFiles, file choosers, MCP file tools) are restricted to workspace root directories (or cwd if no roots are configured).--allow-unrestricted-file-access (and allowUnrestrictedFileAccess config) to explicitly opt out, allowing:
file:// URLs.outputDir/videossaveVideo in configs (https://github.com/microsoft/playwright/pull/38621)return statements.filename parameter. When provided, the snapshot is saved to disk and a link appears in the Files section. This avoids inlining large snapshots and keeps MCP payloads smaller.--console-level=<'error' | 'warning' | 'info' | 'debug'> configure the console severity level that is included in the replies. Defaults to 'info'.--snapshot=<'incremental' | 'full' | 'none'> configure snapshot output mode. Defaults to 'incremental'.{ level: 'error' | 'warning' | 'info' | 'debug' } to include messages up to the given severity level.{ includeStatic: boolean } to include them.Minor fixes and improvements.
Minor fixes and improvements.
We are bringing back the --allow-origins flag! Please make sure you point at the trusted origins only.
There are now multiple ways to provide the initial state to the browser context or a page.
For the storage state, you can either:
--user-data-dir argument. This will persist all browser data between the sessions.--storage-state argument. This will load cookies and local storage from the file into an isolated browser context.For the page state, you can use:
--init-page to point to a TypeScript file that will be evaluated on the Playwright page object. This allows you to run arbitrary code to set up the page. You can use various Playwright APIs there, perform necessary steps, etc.// init-page.ts
export default async ({ page }) => {
await page.context().grantPermissions(['geolocation']);
await page.context().setGeolocation({ latitude: 37.7749, longitude: -122.4194 });
await page.setViewportSize({ width: 1280, height: 720 });
};
--init-script to point to a JavaScript file that will be added as an initialization script. The script will be evaluated in every page before any of the page's scripts.// init-script.js
window.isPlaywrightMCP = true;
There now is a new browser_run_code command that allows LLM to run Playwright APIs as well. LLM can call it with a batch of Playwright APIs:
{
name: 'browser_run_code',
arguments: {
code: `
await page.getByRole("checkbox", { name: "Accept" }).check();
await page.getByRole("button", { name: "Submit" }).click();
`;
}
}
This way LLM can speed up repetitive operations and save on context tokens.
--allowed-origins and --blocked-origins flags are gone. Please use --proxy as a safer alternative or rely upon the underlying OS policies instead.
Fixes WSL client compatibility with Windows servers.
Minor fixes and improvements.
pageState now supports diffs. Unchanged subtrees are collapsed as - ref=<id> [unchanged]. Reduces noise and payload size. microsoft/playwright#37795https://github.com/microsoft/playwright/commit/c813e9cf5 feat(mcp): allow saving videos for sessions https://github.com/microsoft/playwright/issues/37531
# video is saved from the creation of the page to the closure of the context (browser_close call).
npx @playwright/mcp --save-video=800x600
https://github.com/microsoft/playwright/commit/fff065816 feat(mcp): add --init-script option https://github.com/microsoft/playwright/issues/37507
# the code in web-mocks.js will run on page initialization
npx @playwright/mcp --init-script=./web-mocks.js
https://github.com/microsoft/playwright/commit/1313fbd47 chore(mcp): introduce allowed-hosts https://github.com/microsoft/playwright/issues/37541
# Runs MCP server only accessible on 192.168.1.10, defense from DNS rebind
npx @playwright/mcp --allowed-hosts=192.168.1.10:8080 --port=8080
https://github.com/microsoft/playwright/commit/afb59a0ec fix(mcp): tolerate malformed roots https://github.com/microsoft/playwright/issues/37492
https://github.com/microsoft/playwright/commit/0574514a0 feat(mcp): support shared browser context https://github.com/microsoft/playwright/issues/37463
https://github.com/microsoft/playwright/commit/29fb93479 fix(mcp): use single output dir https://github.com/microsoft/playwright/pull/37436
https://github.com/microsoft/playwright/commit/f07c8c0ac chore(mcp): --grant-permissions cli option https://github.com/microsoft/playwright/pull/37431
https://github.com/microsoft/playwright/commit/eb39131eb fix(mcp): --allowed/blocked-origins accept origins not hosts https://github.com/microsoft/playwright/pull/37408
https://github.com/microsoft/playwright/commit/b3d666428 chore: fetch roots lazily https://github.com/microsoft/playwright/pull/37399
https://github.com/microsoft/playwright/commit/4968bc611 chore(mcp): allow to pass extension token via env variable https://github.com/microsoft/playwright/pull/37371
🔥 Secrets 🔥 - https://github.com/microsoft/playwright/commit/dea31d86d
When user prompts
"Authenticate using X-USERNAME / X-PASSWORD"
the secret values for X-USERNAME and X-PASSWORD are not sent to the LLM.
# secrets.env
X-USERNAME=<secret_value>
X-PASSWORD=<secret_value>
Action and navigation timeouts can be configured to go beyond standard 5s/60s. - https://github.com/microsoft/playwright/commit/b37caaaf5
Clicks with keyboard modifiers - https://github.com/microsoft/playwright/commit/cc69e0773
"Shift+click the last todo item"
Cdp connection headers for the browsers in the ☁️ https://github.com/microsoft/playwright/commit/3e4efe9d7
b176111 chore: roll Playwright to latest (#971)
29d468d chore: add verification tools section (#957)
78298c3 chore: introduce verification tools (#951)
7fb8b0d fix: browser_tabs select action with index 0 failing due to falsy check (#964)
7774ad9 chore(extension): support custom executablePath (#947)
3d1a60b chore: introduce form filling tool (#935)
86eba22 chore: remove navigate_forward due to low usage counts (#934)
2521a67 chore: merge tabs tool into one (#933)
Connect to the default Chrome profile using the companion extension
91d5d24 chore: handle list roots in the server, with timeout (#898)
3f148a4 docs: add opencode installation instructions (#895)
53c6b6d fix: backtick quote escaping (#871)
24f81a7 fix: emit code for waitfor (#862)
Minor fixes and improvements.
16f3523 chore: do not return fullPage screenshots to the LLM (#849)
3f6837b fix: cursor does not respond to listRoots (#826)
a60d7b8 chore: slice profile dirs by root in vscode (#814)
2a86ac7 chore: use pngs by default for screenshots (#797)
f600234 chore: record user actions in the session log (#798)
--caps=vision argument) d61aa16--vision mode has been deprecated and turned into --caps=vision capabilitybrowser_pdf_save has been moved into a separate capability, run mcp server with --caps=pdf argument.7fca8f5 chore: roll Playwright to 1.54.1 (#665)
59f1d67 feat(dblclick): add double click (#654)
Minor fixes and improvements.
Minor fixes and improvements.
6567795 chore: respect server settings from config (#502)
eec177d chore: reuse browser in server mode (#495)
54ed7c3 chore: refactor server, prepare for browser reuse (#490)
Beta — feedback welcome: [email protected]