Web Search MCP Server — search tools for AI agents, no API keys

Give Claude, Cursor, n8n, Make — any MCP client — web search, news search, and instant answers in 30 seconds. No Google API keys, no SerpAPI subscription. Hosted on Apify, pay per tool call, failed calls are free.

Tools

Tool	What it returns
web_search	Multi-engine web results: title, url, snippet (region + time filters, pinnable backend)
news_search	Multi-engine news: title, url, source, date, snippet
instant_answer	DuckDuckGo Instant Answers for encyclopedic topics — empty answers are free

Where results come from (honest version): searches run through a metasearch layer that aggregates Bing, Brave, Google, Mojeek, Wikipedia, Yandex and others. backend="auto" (default) blends engines for resilience; pass backend="google", "bing", "mojeek", etc. to pin one (pinned engines can fail when they block datacenter traffic — auto mode is the reliable choice).

Quick start

{
  "mcpServers": {
    "web-search": {
      "url": "https://mrlarryjohnson--web-search-mcp-server.apify.actor/mcp",
      "headers": { "Authorization": "Bearer <your-apify-token>" }
    }
  }
}

Or via Apify's MCP gateway: https://mcp.apify.com?tools=mrlarryjohnson/web-search-mcp-server

• Transport: Streamable HTTP · Auth: Bearer (free Apify account token)
• Pricing: pay per tool call — no subscription, no minimum

Billing you can trust

• Failed calls are never billed — a blocked or rate-limited search raises a clear error instead of returning fake "no results"
• Relevance guard: if the engines return filler that doesn't even contain your query terms, the call raises (unbilled) rather than selling you junk
• Empty instant answers are free — most non-encyclopedic queries have no instant answer; you only pay when there's an actual payload
• Fully async server — concurrent tool calls never block each other

Built by the maker of the Polymarket MCP Server and the whale-tracking suite — same engineering rules throughout.

FAQ

Is this DuckDuckGo? Only the instant_answer tool calls DuckDuckGo's Instant Answer API directly. Web/news search is multi-engine metasearch — that's deliberate: single-engine scraping from datacenter IPs is unreliable, and we'd rather be honest about sourcing than pretend.

What about scraping ethics? Only public results pages/APIs, no logins, no personal data, no CAPTCHA circumvention. Pinned backends that block us simply fail loudly (and free).

More tools by this maker

Tool	What it does
Polymarket Whale Tracker Pro	Insider-scored large trades, top holders, wallet P&L
Polymarket Top Traders	Leaderboards with cross-window consistency + verified P&L
Polymarket MCP Server	The same live data as native AI-agent tools (MCP)
Hyperliquid Whale Tracker	Scored perp trades w/ counterparty addresses, funding/OI, wallet PnL
Web Search MCP Server	Multi-engine web/news search + instant answers for AI agents

Same engineering rules everywhere: fail-loud data integrity, dedup-before-billing, errors never billed.