Google Search MCP Server

Overview

The Google Research MCP Server transforms Google search into a powerful research tool by offering intelligent source ranking, deduplication, agent-based synthesis, and focus area analysis. It's built for Claude Code, Claude Desktop, and other MCP-compatible clients.

Key Features

Available Tools

Quick Reference

Detailed Usage

use_mcp_tool({
  server_name: "google-search",
  tool_name: "research_topic",
  arguments: {
    topic: "WebAssembly performance optimization",
    depth: "basic"
  }
});

use_mcp_tool({
  server_name: "google-search",
  tool_name: "google_search",
  arguments: {
    query: "docker container security best practices",
    num_results: 10,
    dateRestrict: "y1",
    site: "github.com"
  }
});

use_mcp_tool({
  server_name: "google-search",
  tool_name: "extract_webpage_content",
  arguments: {
    url: "https://kubernetes.io/docs/concepts/security/",
    format: "markdown",
    max_length: 5000,
    preview_length: 300
  }
});

Installation

{
  "mcpServers": {
    "google-search": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "mcp-google-search-server"
      ],
      "env": {
        "GOOGLE_API_KEY": "your_google_api_key",
        "GOOGLE_SEARCH_ENGINE_ID": "your_custom_search_engine_id"
      }
    }
  }
}

When to Use This Server

The Google Search MCP server earns its keep when an AI assistant needs current, web-grounded information rather than relying on training data. Three common workflows:

• Research agents — Ask a question and let the assistant run research_topic with multiple searches, deduplicate results, and synthesize a sourced summary instead of a guess.
• Live data checks — Before recommending a library, API, or pricing decision, run a google_search scoped by site or date (site:github.com, dateRestrict: "y1") so the answer reflects reality, not a stale memory.
• Content extraction — Follow search hits with extract_webpage_content to pull the actual page into context, so the assistant reads the source instead of just its title snippet.

It pairs especially well with Fetch MCP for raw URL retrieval and Perplexity MCP for synthesized answers. For an alternative general web-search server, see Brave Search MCP.

Frequently Asked Questions

Sources