HarimxChoi/google-surf-mcp

npx google-surf-mcp   (register in your MCP client config)
Requires Node 18+ and Google Chrome (or Chromium) on the system

Usage

• •Register in ~/.claude.json as {google-surf: {command: 'npx', args: ['-y', 'google-surf-mcp']}}
• •search(query) for ~1.5s results; search_extract(query) returns SERP enriched with ~1500-char abstracts by default
• •On CAPTCHA a Chrome window opens for a human to solve (preserving profile reputation); set SURF_HEADLESS=false to keep Chrome visible

Key features

• ✓5 tools: search, search_parallel (pool of 4), extract, search_extract (search + parallel extract), health
• ✓Drops sponsored ads and knowledge panels (geometric verification); academic PDF extraction from arxiv, biorxiv, Nature, OpenReview, NeurIPS, JMLR, PMLR, Springer, PubMed
• ✓No API key, no proxies, no CAPTCHA solver; Playwright stealth cascade + multi-strategy SERP parser with self-healing strategy reorder
• ✓Fast: ~1.5s/query sequential, parallel x4 ~1.5s; extraction via liteparse (PDFium spatial) + Readability

Best for

AI agents that need reliable, keyless Google search plus article/PDF extraction without paying for search APIs.

Caveats

• ⚠Requires local Google Chrome/Chromium and Node 18+; designed for local use — SURF_CLOUD_MODE=true is fail-fast on CAPTCHA
• ⚠SSRF guard blocks private/loopback addresses in extract by default; rate limit ~10 RPM toward Google