MCP Search Server

A Model Context Protocol (MCP) server that provides web search capabilities via Gemini's grounding API.

Features

• 🔍 Real-time web search powered by Google Gemini
• ⚡ Built-in caching (5 min TTL)
• 🎯 Structured search results (Title + URL + Summary)
• 🌍 Multi-language support

Installation

pip install mcp-search-server

Configuration

Set environment variables:

export MCP_SEARCH_URL="http://localhost:8045/v1/messages"
export MCP_SEARCH_API_KEY="your-api-key"
export MCP_SEARCH_MODEL="gemini-3-flash"  # or gemini-3-pro-high
export MCP_SEARCH_CACHE_TTL="300"  # seconds

Usage

As MCP Server (with Zed, Claude Desktop, etc.)

{
  "mcpServers": {
    "mcp-search": {
      "command": "python",
      "args": ["-m", "mcp_search"],
      "env": {
        "MCP_SEARCH_URL": "http://localhost:8045/v1/messages",
        "MCP_SEARCH_API_KEY": "your-api-key",
        "MCP_SEARCH_MODEL": "gemini-3-flash"
      }
    }
  }
}

As Python Library

import asyncio
from mcp_search import do_search

result = asyncio.run(do_search(
  query="latest AI news",
  model="gemini-3-flash"
))
print(result)

Tool Parameters

Requirements

• Python 3.10+
• Antigravity Manager running locally

License

MIT