Browser automation MCP servers, compared
Browser MCP servers all let an agent open pages, click and read, but they differ in the one thing that decides the fit: whose browser is being driven, and through which protocol. A DevTools server inspects a real Chrome with your logins and a performance panel; a Playwright or Selenium server launches its own clean browser for repeatable tests; a hosted service runs browsers in the cloud and bills per use. This page compares the servers MCP Toplist has checked for this task.
How the 12 servers below were chosen
- The server's documented purpose is driving or inspecting a web browser (not a crawler API alone).
- It has a public repository or published documentation that states its tools and requirements.
- MCP Toplist has read that documentation and written a profile with sources and a check date.
- Ordered by the composite ranking, which measures popularity and activity, not quality — read the cells, not the order.
Cells say what each server's own documentation says; “Not documented” means the docs are silent, not that the feature is absent. Profiles were checked between Oct 9, 2026 and Oct 9, 2026; ranks are as of across 144,396 active servers.
| Server | Engine | Browser session | Best for | Anti-bot / stealth | Requirements | Headless |
|---|---|---|---|---|---|---|
| Chrome DevTools MCP #1chromedevtoolsNo credentials | Puppeteer over Chrome DevTools Protocol | Launches Chrome, or attaches to yours via --autoConnect / --browser-url | Debugging: performance traces, network, console, memory | Not documented | Node.js LTS (npx), Chrome stable; no account | Yes, --headless flag (default headed) |
| Playwright MCP Server #10microsoftNo credentials | Playwright (Chrome, Edge, Firefox, WebKit) | Launches own browser; attaches to Chrome/Edge via extension or --cdp-endpoint | Agent browsing and end-to-end testing via accessibility snapshots | Not documented | Node.js 18+ (npx); no account | Yes, --headless (headed by default); Docker is headless-only |
| Browser Use #11browser-useNo credentials | Browser Use agent on Chrome/Chromium | Launches a local browser; a hosted Cloud MCP server is offered separately | Agent browsing with an LLM-driven fallback | Not documented | Python/uvx, Chrome, your own OpenAI or Anthropic key | Yes, via BROWSER_USE_HEADLESS (set false to show window) |
| SeleniumBase MCP #13seleniumbaseNo credentials | SeleniumBase Pure CDP Mode (Chrome DevTools Protocol, no WebDriver) | Launches its own Chrome; one session at a time | Scraping and browsing on bot-protected sites | Yes: stealthiest CDP mode, solve_captcha for Turnstile, reCAPTCHA, hCaptcha | Python (pip or uv), Chrome/Chromium; no account | Yes; default headless on Linux, headed elsewhere; may be less stealthy |
| Apify MCP Server #14apifyAPI key | Apify Actors (rag-web-browser, web-fetch, Store scrapers) | Hosted on the Apify platform; no local browser | Scraping specific sites with ready-made Actors | web-fetch offers JavaScript rendering and anti-bot protection | Apify account: OAuth or APIFY_TOKEN; npx for stdio | Not documented (Actors run in Apify's cloud) |
| Scrapling MCP Server #15d4vinciNo credentials | Scrapling: browser-fingerprint HTTP, Chromium, and a stealth browser | Launches own browser per call or session; can connect via CDP URL | Scraping, including anti-bot protected pages | Yes: stealth browser bypasses Cloudflare Turnstile/Interstitial; TLS fingerprint impersonation | Python (pip) and scrapling install, or Docker; no account | Yes, headless setting on open_session (ignored for CDP sessions) |
| Firecrawl MCP Server #56firecrawlNo credentials | Firecrawl cloud API (scrape, crawl, interact) | Hosted service; interact sessions persist via scrapeId | Scraping, crawling and web search for agents | Not documented | FIRECRAWL_API_KEY and credits; keyless hosted tier is rate-limited | Not documented (browsers run in Firecrawl's service) |
| Brightdata MCP #107luminati-ioAPI key | Bright Data unlocking network and remote browser | Hosted remote browser; no local setup | Scraping at volume behind bot protection | Yes: bot-detection bypass, CAPTCHA solving, proxy rotation on every request | Bright Data API_TOKEN; free 5,000 requests/month then paid | Remote browser session; headless not stated |
| invisible_playwright_mcp #168feder-crNo credentials | Patched Firefox (invisible_core) driven via Playwright | Launches its own Firefox; optional persistent profile dir | Agent browsing on sites that detect automation | Yes: engine-level fingerprint, seeded identity, proxy-matched locale; no captcha solving | uv/uvx; engine downloads on first start; no account | Yes, default; STEALTHFOX_HEADLESS=0 for headed |
| MCP Server #227browserstackNo credentials | BrowserStack Test Platform (Playwright, Selenium, real devices) | Hosted cloud browsers and real devices | Cross-browser and real-device end-to-end testing | Not documented | BrowserStack account with a plan; OAuth or username + access key | Not documented (sessions run in BrowserStack's cloud) |
| Browser Devtools MCP #442serkan-ozalNo credentials | Playwright (Chromium, Firefox, WebKit); CDP attach for Chromium | Launches own browser; attaches to running Chrome via BROWSER_CDP_ENABLE | Debugging pages and Node.js; package deprecated | Not documented | Node.js 18+ (npx), npx playwright install; no account | Yes, default on (BROWSER_HEADLESS_ENABLE=true) |
| mcp-playwright #1,019executeautomationNo credentials | Playwright (Chromium, Firefox, WebKit) | Launches its own browser per Playwright_navigate | Test codegen and page interaction | Not documented (only a custom user-agent tool) | Node.js (npx); browsers auto-install; no account | Yes, headless parameter on Playwright_navigate (default false) |
- Engine
- Which browser protocol or framework drives the browser?
- Browser session
- Does it launch its own browser, attach to yours, or run in a hosted cloud?
- Best for
- Debugging, end-to-end testing, scraping, or agent browsing?
- Anti-bot / stealth
- Does it document bot-detection avoidance, and how?
- Requirements
- What must be installed, signed in or paid for before first use?
- Headless
- Can it run without a visible window (CI, servers)?
- Need your existing logins, cookies and extensions? Pick a server that attaches to your running Chrome rather than one that launches a fresh profile.
- Writing or debugging end-to-end tests? Prefer the engine your test suite already uses, so selectors and traces transfer.
- Scraping at volume or behind bot protection? Local servers hit rate limits and captchas; hosted services carry that for a fee and their own terms.
- Running in CI or on a server? Confirm headless support and that the server does not need a desktop session.
Each server in detail
- Chrome DevTools MCP →
Chrome DevTools MCP lets a coding agent control and inspect a live Chrome browser through Puppeteer and the DevTools Protocol: record performance traces, read network requests and console messages, take screenshots and automate page actions with automatic waits.
- Playwright MCP Server →
Microsoft's Playwright MCP gives an agent browser automation through Playwright, working from structured accessibility snapshots rather than screenshots, so no vision model is needed and actions apply deterministically.
- Browser Use →
Browser Use's local MCP server exposes its browser agent over stdio so an MCP client can navigate, click, type, read page state and take screenshots, with a fallback tool that hands a whole task to the Browser Use agent when direct control fails.
- SeleniumBase MCP →
The SeleniumBase MCP server provides stealthy browser automation over MCP using SeleniumBase's Pure CDP Mode, which drives Chrome through the DevTools Protocol with no WebDriver in the loop and attempts to handle CAPTCHA challenges.
- Apify MCP Server →
The Apify MCP server lets agents extract data from social media, search engines, maps, e-commerce and other websites by discovering and running Actors from Apify Store, and by managing runs, datasets, tasks and schedules on the Apify platform.
- Scrapling MCP Server →
The Scrapling MCP server brings the Scrapling web-scraping library to agents so they can fetch pages with browser-like HTTP requests, a Chromium browser or a stealth browser that bypasses anti-bot systems, and keep sessions open across calls.
- Firecrawl MCP Server →
Firecrawl MCP Server brings the Firecrawl API to agents: search the web, scrape URLs into structured data or markdown, map and crawl sites, run research agents, monitor pages and interact with pages one turn at a time.
- Brightdata MCP →
Bright Data's MCP server gives agents web search, page scraping, structured data extraction and a remote browser session, with every request routed through Bright Data's unblocking infrastructure for bot-detection bypass, CAPTCHA solving and proxy rotation.
- invisible_playwright_mcp →
invisible_playwright_mcp is an open-source AI browser agent that drives a Firefox patched at the C++ level so the browser looks and behaves like a normal desktop browser; over MCP it exposes tools to open, navigate, read, click and type on real web pages from plain-English instructions.
- MCP Server →
The BrowserStack MCP server connects IDEs and LLM clients to the BrowserStack Test Platform so they can manage test cases, run manual or automated tests on real devices and browsers in BrowserStack's cloud, debug failures and propose fixes.
- Browser Devtools MCP →
Browser DevTools MCP is a Playwright-powered server that gives coding assistants browser automation and debugging (ARIA snapshots, ref-based interactions, screenshots, console and HTTP observability, request stubbing) plus a Node.js debugging platform. The package is marked deprecated and renamed to @ironbee-ai/devtools.
- mcp-playwright →
ExecuteAutomation's Playwright MCP server lets an LLM interact with web pages, take screenshots, scrape visible text and HTML, execute JavaScript in a real browser, and record its actions into generated Playwright test code.