The Playwright MCP server lets an AI agent drive a real browser: click buttons, fill forms, read the page, take screenshots, and inspect network requests. Microsoft builds and maintains it on top of Playwright, the browser automation library already used across a huge share of end-to-end testing in production. Instead of scraping raw HTML or guessing at pixel coordinates, the agent works off an accessibility snapshot of the page, so it can reliably click "the login button" without a vision model or brittle CSS selectors.
That combination, a trusted maintainer plus structured page access instead of screenshots, is why playwright mcp is one of the most searched MCP servers right now. It's also why it's become the default choice for agents that need to test a web app, fill out a multi-step form, or pull data from a site that has no API.
How to Install the Playwright MCP Server
The server ships as an npm package and runs locally over stdio. No API key or account is required to get started.
Claude Code:
claude mcp add playwright npx @playwright/mcp@latest
Claude Desktop, Cursor, Windsurf, VS Code, Zed: add this to your client's MCP config file:
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["@playwright/mcp@latest"]
}
}
}
The first run downloads a Chromium build through Playwright's installer, which can take a minute. After that, the agent has a live browser session it can navigate, click through, and inspect on every subsequent call.
If you only need read access, for example pulling text off pages without letting the agent submit forms or click arbitrary elements, run the server with --caps=core to limit it to navigation and snapshot tools instead of the full interaction set.
Compatible MCP Clients
| Client | Compatible |
|---|---|
| Claude Desktop | Yes |
| Claude Code | Yes |
| Cursor | Yes |
| VS Code | Yes |
| Windsurf | Yes |
| Zed | Yes |
| ChatGPT | No (local stdio servers aren't supported yet) |
Key Tools the Server Exposes
Playwright MCP's tools operate on the page's accessibility tree rather than pixels, which is why actions stay reliable even when a page redesigns its layout. Tools confirmed in the MCPVault listing include:
| Tool | What it does |
|---|---|
browser_navigate |
Load a URL in the browser |
browser_click |
Click an element by reference or description |
browser_hover |
Hover over an element |
browser_drag |
Drag one element onto another |
browser_drop |
Drop a file or data payload onto an element |
browser_fill_form |
Fill multiple form fields in one call |
browser_file_upload |
Attach a file to an upload input |
browser_find |
Search the current page snapshot by text or regex |
browser_handle_dialog |
Accept or dismiss a browser dialog (alert, confirm, prompt) |
browser_evaluate |
Run a JavaScript function against the page or an element |
browser_console_messages |
Read console output, optionally the full session log |
browser_network_requests |
List network requests made since the last navigation |
browser_network_request |
Inspect a single request by index |
browser_navigate_back |
Go back to the previous page |
browser_close |
Close the browser session |
Most tasks don't need a separate screenshot step. The agent reads the accessibility snapshot directly, which is faster and cheaper on tokens than a vision-based screenshot loop, and only falls back to a screenshot when a task specifically needs a visual check.
Quality Grade and Verification Status
Playwright MCP on MCPVault carries a Grade A. MCPVault's methodology weighs maintenance recency, adoption, license clarity and documentation quality. With 36,000+ GitHub stars and a maintainer that ships releases on an active cadence, it clears every signal comfortably. See how MCPVault grades servers for the full scoring breakdown.
The listing is currently unclaimed, meaning it hasn't gone through MCPVault's live handshake verification yet. A Grade A rating reflects strong repository signals. A verified badge would also confirm the server responds correctly to a real MCP connection, re-tested weekly. If Microsoft, or anyone with push access to the repo, claims the listing, verification and a do-follow link become available at no cost during early access. See what a verified badge means for the difference.
If you're looking for an actively developed community fork with different tradeoffs, mcp-playwright by executeautomation is also indexed in the vault with a Grade B and 5,600+ stars. It's worth a look if you want additional convenience helpers layered on top of the core Playwright automation set.
Browse the full Automation category to compare Playwright MCP against other browser and workflow automation servers in the vault.
What Developers Actually Use It For
Three use cases show up repeatedly in how teams connect Playwright MCP.
Agentic QA. Instead of writing a Playwright test script by hand, a developer describes the flow in plain language, "log in, add an item to the cart, check out," and the agent executes it live, catching broken selectors or unexpected redirects as it goes. This doesn't replace a real test suite, but it's a fast way to sanity-check a flow before writing the permanent test.
Scraping sites with no API. A lot of useful data lives behind a web UI with no public API. Because the agent reads the accessibility snapshot rather than parsing raw HTML, it handles dynamic, JavaScript-rendered pages that a simple HTTP request would return empty.
Form-heavy workflows. browser_fill_form fills multiple fields in a single call instead of one tool call per field, which matters for anything with a long intake form, an admin panel, or a multi-step signup flow the agent needs to complete on your behalf.
For all three, the same caution applies: the agent has real click and submit access once connected. Run it with --caps=core when you only need read access, and review what a task is about to do before pointing it at a production account.
Frequently Asked Questions
Does Playwright MCP work with Claude Desktop and Claude Code?
Yes. Claude Code supports it directly through claude mcp add. Claude Desktop needs the JSON config added to its MCP servers file. Both give the agent access to the same tool set once connected.
Do I need to install Playwright separately first?
No. The npx @playwright/mcp@latest command downloads what it needs, including a Chromium browser build, on first run. You don't need an existing Playwright project or a separate playwright install step.
Can I restrict what the agent is allowed to do in the browser?
Yes. Pass --caps=core when launching the server to limit it to navigation and page-reading tools, dropping click, form-fill and file-upload capabilities. This is useful when you want the agent to read a page but not take any action on it.
Is Playwright MCP the same as the executeautomation Playwright server?
No. They're separate projects. Playwright MCP is Microsoft's official server, built directly on the Playwright library it maintains. mcp-playwright by executeautomation is a community project with its own tool set and design choices. Both are indexed on MCPVault with their own grades.
Does Playwright MCP support headless mode?
Yes. Add --headless to the launch args if you don't need to see the browser window, which is typically faster and is the standard choice for CI or background automation tasks.
Your server in the vault.
If you built an MCP server, claim the listing and get it in front of developers who are actively looking. Free to claim. Verification is free during early access and earns a do-follow link to your project.
Claim your listing or submit a server if it is not indexed yet.