The Playwright MCP server provides browser automation capabilities using Playwright through the Model Context Protocol (MCP). Instead of relying on screenshots or visually-tuned models, it operates on Playwright’s accessibility tree, allowing LLMs to interact deterministically with web pages using structured data.

​
When should you use it

Use the Playwright MCP server when you want an agent to:
  • Automate browser actions.
  • Extract structured page context via snapshots, avoiding ambiguity of pixel-based methods.
  • Test and verify UI elements, text, or values without vision models.
  • Capture console logs, network requests, PDFs, or traces during automated workflows.
  • Manage tabs, dialogs, and file uploads in real-time web automation.

​
Requirements

  • Requirements:
    • Node.js 18 or newer
    • An MCP-compatible client (VS Code, Cursor, Windsurf, Claude Desktop, Goose, etc.)
  • Installation:
    Install the Playwright MCP server with your MCP client.
  • Optional capabilities (enabled via --caps):
    • vision → coordinate-based mouse actions
    • pdf → save pages as PDF
    • verify → element/text/value verification
    • tracing → start/stop browser tracing

​
Tools

​
Core interaction

  • browser_click — Click (or double click) on an element.
  • browser_hover — Hover over an element.
  • browser_type — Type text into an editable element, with optional submit/slow typing.
  • browser_fill_form — Fill multiple form fields at once.
  • browser_select_option — Select one or more dropdown values.
  • browser_press_key — Press a keyboard key.
  • browser_drag — Perform drag-and-drop between elements.
  • browser_file_upload — Upload one or multiple files.
  • browser_navigate — Navigate to a specific URL.
  • browser_navigate_back — Go back to the previous page.
  • browser_tabs — List, create, close, or select tabs.
  • browser_close — Close the current page.

​
Page context & capture

  • browser_snapshot — Capture structured accessibility snapshot (preferred for automation).
  • browser_take_screenshot — Take a screenshot of viewport, full page, or element.
  • browser_pdf_save (opt-in via --caps=pdf) — Save page as PDF.

​
Evaluation & debugging

  • browser_evaluate — Run JavaScript in page context.
  • browser_console_messages — Return console messages.
  • browser_network_requests — Return all network requests since load.
  • browser_resize — Resize the browser window.
  • browser_handle_dialog — Accept/decline modal dialogs or prompts.

​
Verification (opt-in via --caps=verify)

  • browser_verify_element_visible — Verify an element is visible by role + accessible name.
  • browser_verify_text_visible — Verify a text string is visible.
  • browser_verify_list_visible — Verify a list with expected items is visible.
  • browser_verify_value — Verify element values (e.g., checkbox state, input value).

​
Coordinate-based (opt-in via --caps=vision)

  • browser_mouse_click_xy — Click at a coordinate.
  • browser_mouse_drag_xy — Drag mouse between coordinates.
  • browser_mouse_move_xy — Move mouse to a coordinate.

​
Tracing (opt-in via --caps=tracing)

  • browser_start_tracing — Start trace recording.
  • browser_stop_tracing — Stop trace recording.

​
Installation

  • browser_install — Install the required browser binaries if not already present.

​
Notes

  • Prefer browser_snapshot over screenshots for interaction—it’s structured, fast, and deterministic.
  • Many tools require both a human-readable element description and a ref from the snapshot for safety and determinism.
  • Optional capabilities (vision, pdf, verify, tracing) must be explicitly enabled when starting the server.