Skip to content
MCP Directory
Back to Directory

Best MCP Servers for Browser Automation

Automate browsers, scrape pages, and interact with web UIs from AI assistants.

Recommended MCP Servers

fetch →everything →filesystem →
COMPARISON OVERVIEW Browser automation sits outside the active official reference server set. Puppeteer-based browser automation was an MCP reference server but now lives in modelcontextprotocol/servers-archived. For live browser control today, use community MCP servers (Playwright-based, Chrome DevTools Protocol wrappers, vendor-specific) discovered via the MCP Registry. ACTIVE OFFICIAL ALTERNATIVE: FETCH Install: uvx mcp-server-fetch Fetch retrieves and converts web content without a headless browser. Choose Fetch when: - Pages are static or server-rendered HTML - You need article text, not button clicks - SSRF and sandbox concerns favor simple HTTP fetch over full browser Limitations: no JavaScript execution, no form submission, no OAuth flows. WHY PUPPETEER WAS ARCHIVED The archived Puppeteer server demonstrated browser automation and scraping patterns. Maintenance moved out of the active reference set — typical for heavy dependencies with large security surfaces. Teams needing clicks, screenshots, and DOM interaction should evaluate actively maintained community servers and run them in isolated environments. FILESYSTEM FOR ARTIFACTS Save screenshots, HAR files, or scraped datasets to disk via Filesystem MCP when automation servers output files. EVERYTHING FOR INTEGRATION TESTS Validate MCP client behavior before trusting browser tools that can exfiltrate logged-in sessions. COMPARISON DIMENSIONS | Dimension | Fetch (official) | Browser automation (community/archived) | | JS rendering | No | Yes | | User login sessions | No | Possible (high risk) | | Maintenance | Active reference | Varies | | Setup | uvx mcp-server-fetch | Often Node + browser binary | SECURITY Browser automation MCP is high risk: cookies, MFA bypass attempts, arbitrary navigation. Never connect a logged-in browser profile to an LLM without strict tool approval. Prefer Fetch for documentation reading. Use dedicated automation profiles with no corporate SSO. DECISION GUIDE Static docs / blogs → Fetch. SPAs and dashboards → community Playwright MCP in sandbox. Legacy Puppeteer reference → study archived code only, do not deploy unpatched.