The best Browser automation MCP servers
Give an agent a browser and the web stops being a wall. These servers split into two jobs that get conflated: driving a real browser — clicking, filling, screenshotting, reading the console — and extracting content at scale from pages you do not control. Pick by the job, because a scraper cannot log in for you and a driver will not survive ten thousand pages.
Official connectors first, then verified entries, then by how many tools they expose.
Playwright MCP
microsoft
Browser automation with Playwright. Navigate pages, fill forms, click elements with cross-browser support.
Puppeteer
anthropic
Browser automation and web scraping using Puppeteer. Navigate pages, take screenshots, and interact with web elements.
Chrome DevTools
Drive a real Chrome instance from an agent: navigate, inspect the DOM, read console and network activity, and capture performance traces.
Browserbase
browserbase
Cloud browser automation using Stagehand and Browserbase, enabling LLMs to interact with web pages, take screenshots, and run parallel sessions.
BrowserTools MCP
AgentDeskAI
Capture browser logs, network activity, screenshots, and interact with browser DevTools from AI agents.
Firecrawl
mendableai
Advanced web scraping that handles JavaScript-rendered pages, removes boilerplate, and outputs clean markdown.
Apify
apify
Web scraping and automation platform. Run actors for data extraction, web crawling, and browser automation.
Bright Data
brightdata
Web data collection platform. Scrape any website with residential proxies and SERP API.
Compare all 8 at a glanceShowHide
| Server | Tools | Status |
|---|---|---|
| Playwright MCP | 8 | — |
| Puppeteer | 8 | Archived |
| Chrome DevTools | 5 | Verified |
| Browserbase | 7 | — |
| BrowserTools MCP | 6 | — |
| Firecrawl | 6 | Verified |
| Apify | 7 | — |
| Bright Data | 6 | — |
How we rank theseShowHide
Official connectors we host come first, then verifiedentries — ones where we hold the server’s own repository or the vendor’s own docs and have checked the link resolves — then the rest by how many agent-callable tools they expose. A listing on another directory doesn’t count as verification. We don’t publish install counts for third-party servers, because we have no way to measure them, and anything its maintainers have archived says so on the card. A dash in the Tools column means we haven’t listed that server’s tools yet, not that it has none.
Frequently asked questions
What is a browser automation MCP server?
It exposes a browser to an AI model through the Model Context Protocol. The model gets tools like navigate, click, type and screenshot, so it can work through a site the way a person would — useful for anything with no API, and for testing your own product end to end.
Driving a browser or scraping — which do I need?
Driving, if the task needs a session: logging in, filling a form, stepping through a checkout, reproducing a bug. Scraping, if you need the content of many pages and not the interaction — that is where a managed extractor earns its cost, because it handles rotation, rendering and rate limits you would otherwise build.
Do these run locally or in the cloud?
Both. Playwright, Puppeteer and Chrome DevTools drive a browser on your own machine, which keeps sessions and cookies local. Browserbase, Apify and Bright Data run it on their infrastructure, which is what you want for scale or for a fixed egress IP. Each listing says which.
Is scraping with an AI agent legal?
The same rules apply as to any crawler: respect robots.txt, honour the site’s terms, do not collect personal data without a basis, and do not hammer a server. An agent is faster than a person, which makes restraint a setting you have to configure rather than a side effect of typing speed.
Browse other categories
Connect Playwright MCP through one endpoint
Reach Playwright MCP and the rest of this list from Claude, ChatGPT, Gemini or Cursor without configuring each one. Free plan, no credit card.
Get started freeOr read how managed MCP hosting works first.