Playwriter

Let's be brutally honest: most AI browser automation tools are a pain. They either lock your agent in a sterile, cookie-less sandbox that gets flagged as a bot instantly, or they give it a clunky, limited set of pre-defined "tools" that can't handle real-world web complexity. Playwriter is the antidote. It's a Chrome extension and CLI that hands the full power of the Playwright automation API directly to your AI agent, but with one critical twist: it runs inside your actual, logged-in browser session. This means your AI can navigate the web with all your extensions, cookies, and saved logins already in place, bypassing the instant detection that plagues headless instances. It's like giving your AI a driver's license for your personal browser. Built as an open-source MCP (Model Context Protocol) server, it integrates seamlessly with clients like Cursor, Claude Desktop, and VS Code. The philosophy is powerful yet simple: one single execute tool that can run any Playwright code, eliminating schema bloat and giving developers and AI agents unprecedented, granular control over the browsing experience. This isn't just automation; it's a collaborative browsing session where the AI handles the tedious work, and you step in as the human-in-the-loop when needed.

Your Actual Browser Session

This is the killer feature. Instead of spawning a fresh, suspicious Chrome instance, Playwriter attaches directly to your existing browser via a Chrome extension. Your agent operates in an environment with all your personal logins, saved cookies, installed extensions (like ad-blockers or password managers), and even your existing tabs. This eliminates bot detection triggers from "new" browser fingerprints and avoids the memory overhead of running a second Chrome process. It’s the difference between sending a stranger to do your online shopping versus giving your trusted friend your laptop.

Full Playwright API via a Single Tool

Forget menus of dozens of limited, pre-wrapped tools like "click_button" or "extract_text." Playwriter exposes one powerful execute command that accepts any valid Playwright code. This means your AI agent can use the entire, mature Playwright API for complex interactions: waiting for specific network responses, setting JavaScript breakpoints, manipulating localStorage, or taking efficient accessibility snapshots. This approach drastically reduces context window usage and provides maximum flexibility, trusting the developer or AI to write precise instructions.

Built-in Debugger & Live Editor

Playwriter transforms your browser into a debuggable runtime for AI agents. You can set breakpoints in the agent's Playwright script, pause execution, and inspect the state of the page. The live code editing feature allows you to modify the agent's commands on-the-fly without restarting the entire session. This is invaluable for troubleshooting, refining scripts, or when the agent gets stuck on an unexpected UI element. It brings a developer-grade workflow to AI-powered automation.

Lightweight Accessibility Snapshots

Traditional full-page screenshots are context hogs, often exceeding 100KB. Playwriter's accessibility snapshots are a game-changer, typically weighing only 5-20KB. They capture the semantic structure of the page—roles, names, states, and relationships—which is exactly what an AI agent needs to understand the page layout and interact with elements. This makes agent interactions faster, more accurate, and far cheaper in terms of token usage compared to processing large image files.

AI-Assisted Web Research & Data Extraction

Need to compile a report from multiple sources that require login? Instruct your AI agent to navigate your logged-in news subscriptions, academic journals, or business intelligence platforms. It can click through pagination, handle consent modals with your guidance, and extract structured data—all within your authenticated session, avoiding paywalls and login barriers that stop other automation tools cold.

Automated Testing with Real User Context

Developers can use Playwriter to create and run integration tests that mimic real user journeys. Since it uses your actual browser profile, tests can run against staging environments that require specific authentication cookies or against complex workflows that depend on browser extensions. This provides a more accurate testing environment than isolated, clean browser instances.

Routine Administrative Task Automation

Automate the tedious, repetitive web tasks that clutter your day. This could be filling out recurring forms, checking statuses on multiple dashboards, downloading daily reports from a web portal, or managing content across platforms like Shopify or WordPress. Your AI handles the routine clicks and inputs, and you simply supervise or step in for exceptions like CAPTCHAs.

Collaborative Browsing & Pair Programming

This is where Playwriter shines as a collaboration tool. You can watch the AI navigate in real-time on your screen. When it encounters a hurdle—a tricky multi-factor authentication step, an ambiguous "I'm not a robot" checkbox, or a novel UI—you can immediately intervene. Disable the extension on that tab, solve the human-required problem manually, re-enable control, and let the AI continue. It's true human-AI teamwork.

Is my browsing data sent to a remote server?

Absolutely not. Playwriter is designed with privacy first. The architecture is local: the Chrome extension connects to a WebSocket relay running on localhost:19988 on your own machine. Your AI client (CLI, MCP) also connects to this local relay. All commands and data (CDP traffic) flow directly between your browser and your local client. No data is sent to any remote server, and no account is required.

How does it avoid bot detection?

It avoids the classic red flags of automation. Because it uses your existing, long-lived Chrome session with a normal history of use, cookies, and extensions, it presents a browser fingerprint that looks entirely human. Websites see it as you browsing, not a fresh, sterile automation environment. The extension uses the official chrome.debugger API, which is a supported method for development tools.

Can I use it with any AI or just specific clients?

Playwriter is built on the open Model Context Protocol (MCP), making it client-agnostic. It works seamlessly with any MCP-compliant client. This includes popular AI-powered editors like Cursor and Windsurf, desktop agents like Claude Desktop, and code editors like VS Code with an MCP plugin. The provided CLI also lets you drive it directly from your terminal or your own scripts.

What happens if the AI gets stuck or makes a mistake?

You have full control. You can see every action happening live in your browser. If the agent enters a loop or starts clicking the wrong thing, you can instantly click the extension icon to detach it (turning it gray) and regain manual control of the tab. After you fix the state of the page, re-attach the extension, and the agent can pick up from there. The built-in debugger also allows you to pause and step through the agent's commands.