Browser Automation MCP Servers Compared (2026): Playwright, Browserbase, Browserless, Browser Use, Bright Data and Chrome DevTools

• Playwright MCP (Microsoft): local browser control with tools like browser_navigate, browser_click and browser_snapshot. Runs on your machine. Free and open source (Apache License 2.0). No paid plan.
• Chrome DevTools MCP (Google): local Chrome control plus performance traces, network and console inspection, Lighthouse audits and memory snapshots. Runs on your machine. Free and open source (Apache License 2.0). No paid plan.
• Browserbase MCP: cloud browsers driven by natural-language tools (navigate, act, observe, extract), built on Stagehand. Hosted endpoint or local npx. Free: 1 browser hour and 3 concurrent browsers, no CAPTCHA solving. Paid: Developer, $20/month for 100 browser hours.
• Browserless MCP: cloud browsers with a stateful browserless_agent tool plus scrape, search, crawl, export and performance tools. Hosted endpoint or local npx. Free: 1k units/month and 2 concurrent browsers, with CAPTCHA solving. Paid: Prototyping, $25/month billed annually for 20k units.
• Browser Use MCP: hosted endpoint; you send a task in plain English and a Browser Use agent runs it in a cloud browser (run_session, send_task). Free: a one-time $1 credit for eligible signups, no subscription. Paid: pay as you go, with no monthly plan; cloud browsers cost $0.02 per browser-hour plus model tokens.
• Bright Data browser tools: 13 scraping_browser_* tools that drive a remote browser through Bright Data's unblocking network. Hosted endpoint or local npx. Free: 5,000 requests/month, no credit card, browser automation included. Paid: pay as you go, $8/GB for browser navigation; the Scale plan is $499/month.
• Prices and free tiers were checked on each vendor's pricing or docs page on 2026-10-23 and change often.

Playwright MCP runs a real browser on your machine. It reads pages through Playwright's accessibility tree instead of screenshots, so the model does not need vision. The browser opens with a visible window by default; add --headless for servers and CI.

claude mcp add playwright npx @playwright/mcp@latest

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

• Main tools: browser_navigate, browser_snapshot, browser_click, browser_type, browser_fill_form, browser_select_option, browser_take_screenshot, browser_wait_for, browser_tabs, browser_network_requests, browser_console_messages, browser_evaluate.
• Options: --headless, --browser (chrome, firefox, webkit, msedge), --isolated (keep the profile in memory), --user-data-dir, --caps (vision, pdf, devtools), --cdp-endpoint (connect to an existing browser), --port.
• Each option has a matching env var, such as PLAYWRIGHT_MCP_HEADLESS, PLAYWRIGHT_MCP_BROWSER and PLAYWRIGHT_MCP_CDP_ENDPOINT.
• Gotcha: browser_run_code_unsafe is in the core tool set and runs arbitrary code in the Playwright server process; the README calls it RCE-equivalent. Use it only with prompts you trust.
• Gotcha: the README says coding agents increasingly favor Playwright's CLI with skills over MCP; MCP still fits loops that need persistent state.
• Requires Node.js 18 or newer. No cloud browsers, proxies or CAPTCHA solving.

• Playwright MCP server: install for every agent

Chrome DevTools MCP gives the agent a Chrome browser plus DevTools: performance traces, network requests, console messages, Lighthouse audits and heap snapshots. Pick it to debug or speed up your own site, not to scrape other sites.

claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latest

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp@latest"]
    }
  }
}

• Main tools: navigate_page, new_page, click, fill, fill_form, type_text, take_snapshot, take_screenshot, evaluate_script, list_network_requests, list_console_messages, performance_start_trace, performance_stop_trace, lighthouse_audit, take_heapsnapshot.
• Options: --headless, --isolated, --slim (a smaller tool set), --no-usage-statistics, --no-performance-crux.
• Gotcha: usage statistics are on by default. Turn them off with --no-usage-statistics or the CHROME_DEVTOOLS_MCP_NO_USAGE_STATISTICS env var.
• Gotcha: performance tools may send trace URLs to the Google CrUX API; use --no-performance-crux to stop that.
• Officially supports Google Chrome and Chrome for Testing only. Needs Node.js LTS and current stable Chrome.

Browserbase runs the browser in its cloud. Its MCP server is built on Stagehand, so tools take plain-English instructions such as "click the login button" instead of selectors. Browserbase recommends the hosted endpoint.

claude mcp add --transport http browserbase "https://mcp.browserbase.com/mcp?browserbaseApiKey=YOUR_BROWSERBASE_API_KEY"

{
  "mcpServers": {
    "browserbase": {
      "command": "npx",
      "args": ["@browserbasehq/mcp"],
      "env": {
        "BROWSERBASE_API_KEY": "YOUR_BROWSERBASE_API_KEY",
        "GEMINI_API_KEY": "YOUR_GEMINI_API_KEY"
      }
    }
  }
}

• Main tools: start, navigate, act, observe, extract, end. Each accepts an optional sessionId.
• Env vars for local use: BROWSERBASE_API_KEY and GEMINI_API_KEY (the model Stagehand uses).
• Local flags listed in the GitHub README: --proxies, --verified (--advancedStealth is a deprecated alias), --keepAlive, --contextId, --persist, --modelName, --modelApiKey.
• Gotcha: the GitHub repo for the local server is archived and no longer maintained. Use the hosted endpoint.
• Gotcha: the free plan has no CAPTCHA solving and no stealth. Auto CAPTCHA solving and 1GB of proxies start on the $20 Developer plan.

• Browserbase MCP server: install for every agent

Browserless runs browsers in its cloud. Its MCP server mixes one stateful agent tool with stateless REST tools for scraping, search, crawling and exports. The hosted endpoint accepts a Bearer token, a token query parameter, or OAuth sign-in.

claude mcp add --transport http browserless https://mcp.browserless.io/mcp \
  --header "Authorization: Bearer YOUR_API_TOKEN_HERE"

{
  "mcpServers": {
    "browserless": {
      "command": "npx",
      "args": ["-y", "@browserless.io/mcp"],
      "env": { "BROWSERLESS_TOKEN": "YOUR_API_TOKEN_HERE" }
    }
  }
}

• Main tools: browserless_agent (stateful browser), browserless_smartscraper, browserless_function, browserless_export, browserless_search, browserless_map, browserless_crawl, browserless_performance, browserless_skill, plus read-only account tools such as browserless_usage and browserless_sessions.
• Env vars: BROWSERLESS_TOKEN (required), BROWSERLESS_API_URL (default https://production-sfo.browserless.io; London and Amsterdam endpoints exist), TRANSPORT, PORT, BROWSERLESS_TIMEOUT, BROWSERLESS_MAX_RETRIES, BROWSERLESS_CACHE_TTL.
• Gotcha: billing is in units. A unit is up to 30 seconds of browser time per connection. Residential proxies cost 6 units per MB and each solved CAPTCHA costs 10 units.
• Gotcha: all paid prices on the pricing page are billed annually (Prototyping $25/month, Starter $140/month, Scale $350/month); monthly-billing prices are not shown.

• Browserless MCP server: install for every agent

Browser Use's MCP server is a hosted endpoint at api.browser-use.com. Instead of clicking step by step, the agent hands a whole task to a Browser Use agent, which runs it in a cloud browser and reports back. It runs API V3 sessions.

claude mcp add -t http -H "Authorization: Bearer YOUR_API_KEY" browser-use https://api.browser-use.com/v3/mcp

{
  "mcpServers": {
    "browser-use": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.browser-use.com/v3/mcp",
        "--header",
        "Authorization:${BROWSER_USE_AUTH_HEADER}"
      ],
      "env": { "BROWSER_USE_AUTH_HEADER": "Bearer YOUR_API_KEY" }
    }
  }
}

• Main tools: run_session, send_task, get_session, get_session_messages, stop_session, list_sessions, list_browser_profiles.
• Auth: Authorization: Bearer YOUR_API_KEY. The older x-browser-use-api-key header still works.
• Gotcha: V3 uses token-based pricing: 1.2x provider token rates with a Browser Use-hosted model, or provider cost plus a 0.2x fee with your own model key. Cloud browsers add $0.02 per browser-hour.
• Gotcha: the free $1 credit is one-time and does not renew.

• Browser Use MCP server: install for every agent

Bright Data's MCP server includes a browser group of 13 scraping_browser_* tools. They drive a remote browser through Bright Data's network, which is built for sites that block bots. To load them, add groups=browser to the hosted URL or set GROUPS=browser locally.

claude mcp add --transport http brightdata "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN&groups=browser"

{
  "mcpServers": {
    "Bright Data": {
      "command": "npx",
      "args": ["@brightdata/mcp"],
      "env": {
        "API_TOKEN": "YOUR_API_TOKEN",
        "GROUPS": "browser"
      }
    }
  }
}

• Browser tools: scraping_browser_navigate, scraping_browser_go_back, scraping_browser_go_forward, scraping_browser_snapshot, scraping_browser_click_ref, scraping_browser_type_ref, scraping_browser_screenshot, scraping_browser_get_text, scraping_browser_get_html, scraping_browser_scroll, scraping_browser_scroll_to_ref, scraping_browser_wait_for_ref, scraping_browser_network_requests.
• Env vars: API_TOKEN (required), BROWSER_ZONE (default mcp_browser), GROUPS, TOOLS (add single tools).
• Gotcha: browser navigation is billed by bandwidth ($8/GB pay as you go, $6/GB on Scale), while search and scrape are billed per request ($1.5 per 1K).

• Bright Data MCP server: install for every agent

• Local testing while you code: Playwright MCP. Add Chrome DevTools MCP when you need performance traces or Lighthouse.
• CI: Playwright MCP with --headless and --isolated, so no profile is left on disk and no account is needed.
• Scale and stealth: Browserbase (stealth and proxies on paid plans, advanced stealth on Scale) or Browserless (regional endpoints in the US and Europe, with CAPTCHA solving billed in units).
• Sites that block bots: Bright Data's browser tools, or Browserless, which includes CAPTCHA solving and proxies in its free plan.
• Natural-language tasks: Browser Use for whole tasks from one instruction, or Browserbase's act and extract for step-by-step plain-English control.
• Prices and free tiers were checked on each vendor's pricing or docs page on 2026-10-23.

• Set up MCP servers, rules and permissions in one paste
• Add an MCP server in every AI coding agent
• Web scraping MCP servers compared