You use and automate browsers by running playwright-cli commands through the terminal MCP server's run_command tool. Pass each command as the command argument to run_command. Always run the request_session_mode command first to get approval for multiple commands in a row without user prompts.

Setup

Before first use, install the CLI globally:

npm install -g @playwright/cli

This provides the playwright-cli binary. Do NOT confuse with npx playwright open — that is a different tool (interactive codegen) that blocks the terminal.

How It Works

playwright-cli is a daemon-based CLI (NOT a TUI). It works like this:

playwright-cli open spawns a background browser daemon, opens the browser, and returns with a snapshot. The command exits — it does NOT block.
Subsequent commands (click, fill, snapshot, etc.) talk to the daemon, print output, and exit.
playwright-cli close stops the daemon and closes the browser.

Because each command is short-lived: use regular run_command — no PTY or background mode needed.

Headed vs headless: The browser is headless by default (invisible). Always pass --headed to open so the user can see the browser:

playwright-cli open [https://example.com](https://example.com) --headed

Use playwright-cli show to open a visual dashboard of all running sessions.

Critical Workflow

Execute via run_command — pass each playwright-cli command as the command argument. Never suggest the user paste commands manually.
Read the snapshot via read_file — after open, goto, click (that navigates), or calling playwright-cli snapshot, a page snapshot is generated. You must call the read_file MCP tool to read the snapshot before moving on to your next command. This file contains the current URL, page title, and an accessibility tree of interactive elements labeled with ref IDs (e.g. e1, e5, e12). You MUST use these ref IDs for subsequent interactions.
Snapshot after navigation — run playwright-cli snapshot any time you need a fresh view or after a page change, then immediately follow up with read_file to parse the state.
One command at a time — run a single playwright-cli command per run_command call, use read_file to check the output, then decide the next action. Never chain multiple commands.
Always close when done — playwright-cli close or playwright-cli close-all.

Quick Start

Each line below is a separate step in your workflow:

playwright-cli open [https://example.com](https://example.com) --headed
# → Call `read_file` to read the snapshot output: e1 [link "More information..."], etc.

playwright-cli click e1
# → Call `read_file` to read the new snapshot after navigation

playwright-cli snapshot
# → Explicitly get current page state if needed
# → Call `read_file` to read the fresh snapshot

playwright-cli close

Typing into Inputs & Search Bars

This is the most common source of errors for small models. Follow this exact pattern for any input field on any page:

Step 1: Open the page and read the snapshot carefully using read_file.

playwright-cli open [https://example.com](https://example.com) --headed

Read the file containing the snapshot with element refs. Find the input you need — look for [searchbox], [combobox], [textbox], or [input] entries with a ref ID.

Step 2: Click the input first to focus it, then fill it.

playwright-cli click e5
# → e5 is the ref ID of the input found in the snapshot

Step 3: Fill the input with your text. Use fill (NOT type) — it clears the field first.

playwright-cli fill e5 "your text here"

Step 4: Press Enter to submit, or click a submit button if one exists.

playwright-cli press Enter
# OR: playwright-cli click e8   (if there's a submit/search button)

Key rules for inputs:

Always use the ref ID from the snapshot — never guess or hardcode a ref.
fill <ref> "text" targets a specific input field (clears first, then types). Use this for any search bar, form field, text input, etc.
type "text" types into whatever is currently focused — less reliable. Avoid unless fill does not apply.
If a snapshot doesn't show the element you expect, call playwright-cli snapshot again and read the new file — the page may still be loading dynamically.

Understanding Snapshots

When you use read_file after a command, the snapshot structure looks like:

### Page
- Page URL: [https://example.com/](https://example.com/)
- Page Title: Example Domain
### Snapshot
- [document] Example Domain
  - [heading] "Example Domain"
  - [paragraph] "This domain is for use in illustrative examples..."
  - [link e1] "More information..."

The ref IDs (e1, e2, …) are your handles for interacting with elements. Always read the snapshot file to find the correct ref before clicking, filling, or interacting.

Command Reference

Open & Close

playwright-cli open --headed                  # open browser (visible to user)
playwright-cli open [https://example.com](https://example.com) --headed  # open and navigate
playwright-cli open --browser=chrome --headed # specific browser (chrome/firefox/webkit/msedge)
playwright-cli open --persistent --headed     # persistent profile on disk
playwright-cli show                           # open visual dashboard of all sessions
playwright-cli close                          # close browser
playwright-cli close-all                      # close all sessions
playwright-cli kill-all                       # force-kill all processes

Navigation

playwright-cli goto [https://playwright.dev](https://playwright.dev)
playwright-cli go-back
playwright-cli go-forward
playwright-cli reload

Element Interaction (use ref IDs from snapshot)

playwright-cli click e3                       # click
playwright-cli dblclick e7                    # double-click
playwright-cli fill e5 "user@example.com"     # clear + type into input
playwright-cli type "search query"            # type into focused element
playwright-cli select e9 "option-value"       # select dropdown
playwright-cli check e12 / uncheck e12        # checkbox
playwright-cli hover e4                       # hover
playwright-cli drag e2 e8                     # drag and drop
playwright-cli upload ./document.pdf          # file upload

Snapshots & Screenshots

playwright-cli snapshot                       # generate accessibility tree (read via read_file next!)
playwright-cli screenshot                     # full page screenshot
playwright-cli screenshot e5                  # element screenshot
playwright-cli screenshot --filename=page.png # named screenshot
playwright-cli pdf --filename=page.pdf        # PDF export

Keyboard & Mouse

playwright-cli press Enter
playwright-cli press ArrowDown
playwright-cli keydown Shift / keyup Shift
playwright-cli mousemove 150 300
playwright-cli mousedown / mouseup
playwright-cli mousewheel 0 100

JavaScript, Dialogs & Window

playwright-cli eval "document.title"
playwright-cli eval "el => el.textContent" e5
playwright-cli dialog-accept / dialog-dismiss
playwright-cli resize 1920 1080

Tabs

playwright-cli tab-list
playwright-cli tab-new [https://example.com](https://example.com)
playwright-cli tab-select 0
playwright-cli tab-close 2

Named Sessions (parallel isolated browsers)

playwright-cli -s=auth open [https://app.example.com/login](https://app.example.com/login) --headed
playwright-cli -s=auth fill e1 "user@example.com"
playwright-cli -s=auth close
playwright-cli list                           # list all sessions

Storage, Network & DevTools (see references for details)

playwright-cli state-save auth.json / state-load auth.json
playwright-cli cookie-list / cookie-set name value
playwright-cli localstorage-get key / localstorage-set key value
playwright-cli route "**/api/*" --body='{"mock":true}' / unroute
playwright-cli console / network
playwright-cli tracing-start / tracing-stop
playwright-cli video-start / video-stop out.webm
playwright-cli run-code "async page => { ... }"

Reference Docs

For advanced usage, read the reference files in the references/ subfolder:

references/request-mocking.md — intercept & mock network requests
references/running-code.md — run arbitrary Playwright code
references/session-management.md — parallel browser sessions
references/storage-state.md — cookies, localStorage, sessionStorage
references/test-generation.md — generate Playwright test files
references/tracing.md — capture execution traces
references/video-recording.md — record browser sessions