MCPRisingofficial publisher

Playwright MCP

MCP : Standardises the connection between an AI application and systems that expose tools or data · requires tout client MCP

Gives the agent a controllable browser: read pages, click, fill forms, test.

The reference navigation MCP, maintained by Microsoft.

GitHub · 38,600 ★

Overall score

8.0 /10

Security: one caveat, see the card

Utility

8.5

Reliability

10.0

Security

One caveat

Ease

5 min · Comfortable

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Controlling a browser from an AI agent usually requires screenshots and vision-capable models, which is slow and costly. Playwright MCP solves this by exposing the accessibility tree of web pages, a structured representation of content, directly readable by a language model without any vision capability.

In practice, the MCP server is installed locally via Node.js and connects to your MCP client (VS Code, Cursor, Claude Code, etc.). The agent can then open pages, click elements, fill forms and read content in a deterministic way, without the ambiguity of visual interpretation.

This tool is aimed at developers who want to automate web tasks, write tests or build autonomous workflows requiring continuous browser interaction, particularly for exploration, self-healing tests or long-running processes.

It is less suited to high-throughput coding agents that juggle between navigation and large codebases: in that case, Microsoft recommends Playwright CLI with the SKILLS system instead, which is more token-efficient.

What it offers

·Agent-driven web navigation: opening pages, clicking, filling forms and reading content.
·Uses the accessibility tree rather than screenshots, requiring no vision model.
·Deterministic action application, without ambiguity from visual interpretation.
·Support for multiple browsers: Chrome, Firefox, WebKit and Microsoft Edge.
·Options to restrict allowed origins and hosts to limit the domains accessible by the agent.
·Ability to block service workers and control access to local files.
·Connection to an existing CDP endpoint to control an already open browser.
·Optional capabilities that can be enabled: vision, PDF, developer tools.

Our tips

Use the --browser option to choose the right browser for your use case (chrome, firefox, webkit, msedge) depending on the sites to automate.
Enable optional capabilities with --caps vision,pdf,devtools only when your workflow requires them, to avoid unnecessarily loading the model context.
Use --blocked-origins or --allowed-origins to restrict the domains the agent can contact, reducing the risk of unwanted actions.
To control an already open browser (with your active session), use --cdp-endpoint rather than letting the server open a new one.
If you are working on a high-throughput coding agent, evaluate Playwright CLI with SKILLS: it is more token-efficient as it avoids loading large tool schemas into the context.

Watch out

The agent acts with your active browser sessions: avoid using it while logged into sensitive accounts (banking, professional email, etc.).
For coding agents with large context windows, the MCP can become token-costly due to loading the accessibility tree; Playwright CLI with SKILLS is better suited in that case.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: claude-code, cursor, vscode
Host: tout client MCP
Licence: Apache-2.0

Security and data

Rating: Security: one caveat, see the card
Your data: Data stays local
Maintenance: Active
Last release: 28 September 2026

Links and repository checked on 25 October 2026 · added on 1 March 2025

How to install Playwright MCP

1. Make sure Node.js 18 or newer is installed on your machine. 2. Open the MCP settings of your client (VS Code, Cursor, Claude Code, etc.). 3. Add the following configuration: {"mcpServers":{"playwright":{"command":"npx","args":["@playwright/mcp@latest"]}}}. 4. Save and restart your client if needed. 5. Caution: do not use this server while logged into sensitive accounts, as the agent can act on the web using your active sessions.

npx @playwright/mcp@latest

In the same category

Whole category Feed (memory, context) →