Efficient Search

Part of the Efficient MCP series - Web search optimized for LLMs with intelligent caching and rich context.

🎉 Now Published! Install with bunx @om-surushe/efficient-search

Quick Start

# Install globally
npm install -g @om-surushe/efficient-search

# Or run directly
bunx @om-surushe/efficient-search

Get your Google PSE credentials:

1. Create a Programmable Search Engine: https://programmablesearchengine.google.com/
2. Get API key: https://console.cloud.google.com/apis/credentials
3. Copy Search Engine ID from PSE dashboard

Add to your MCP client config:

{
  "mcpServers": {
    "efficient-search": {
      "command": "bunx",
      "args": ["@om-surushe/efficient-search"],
      "env": {
        "GOOGLE_API_KEY": "your_api_key_here",
        "SEARCH_ENGINE_ID": "your_search_engine_id"
      }
    }
  }
}

What Makes This Different?

Traditional Google Search API:

{
  "title": "Example Page",
  "link": "https://example.com",
  "snippet": "Some text..."
}

→ LLM has to extract metadata, clean HTML entities, rank results = More tokens, slower

Efficient Search MCP:

{
  "title": "Example Page",
  "url": "https://example.com",
  "snippet": "Clean, formatted text",
  "displayUrl": "example.com",
  "relevance": 0.95,
  "metadata": {
    "description": "Full page description",
    "author": "Author name",
    "publishedDate": "2024-01-01",
    "thumbnail": "https://...",
    "siteName": "Example"
  },
  "summary": "Found 1,234 results. Most relevant: ..."
}

→ Everything pre-processed. LLM just reads and responds. = Faster, efficient, cached

Features

• 🔍 Clean, structured results - No HTML entities, formatted for LLM consumption
• 🧠 Rich metadata extraction - Author, publish date, thumbnails, descriptions
• ⚡ Smart caching - 60min default TTL, configurable
• 🎯 Relevance scoring - Pre-calculated relevance for each result
• 📊 LLM-friendly summaries - "Found X results, most relevant: ..."
• 🌍 Geolocation & language - Filter by country and language
• 🔒 Safe search - Configurable safety levels
• 🚀 Built with Bun - Fast, modern TypeScript runtime

Available Tools

Tool	Description
web_search	Search the web with LLM-optimized results
clear_cache	Clear cached search results
get_cache_stats	View cache size, TTL, and hit rate

Configuration

Environment variables:

Variable	Required	Default	Description
GOOGLE_API_KEY	✅	-	Google Cloud API key
SEARCH_ENGINE_ID	✅	-	Programmable Search Engine ID
CACHE_TTL_MINUTES	❌	60	Cache time-to-live in minutes
MAX_RESULTS	❌	10	Maximum results per query

Usage Examples

Basic Search

web_search({ query: "typescript best practices" })

Advanced Search with Filters

web_search({
  query: "machine learning papers",
  num: 5,
  gl: "us",
  lr: "lang_en",
  safe: "high"
})

Response Format

{
  "query": "typescript best practices",
  "totalResults": 12400000,
  "searchTime": 0.45,
  "cached": false,
  "summary": "Found 12,400,000 results. Most relevant: ...",
  "results": [
    {
      "title": "TypeScript Best Practices Guide",
      "url": "https://example.com/guide",
      "snippet": "Clean, formatted snippet...",
      "displayUrl": "example.com",
      "relevance": 1.0,
      "metadata": {
        "description": "Comprehensive guide...",
        "author": "John Doe",
        "publishedDate": "2024-01-15"
      }
    }
  ]
}

Tech Stack

• Runtime: Bun - Fast JavaScript runtime
• Language: TypeScript 5.3+
• Protocol: Model Context Protocol (MCP)
• API: Google Programmable Search Engine

Development

# Install dependencies
npm install

# Run in dev mode
bun run dev

# Build
bun run build

# Type check
bun run typecheck

# Lint
bun run lint

Part of Efficient MCP Series

• @om-surushe/efficient-ticktick - LLM-optimized TickTick task management
• @om-surushe/efficient-search - LLM-optimized web search (this package)
• More coming soon...

All packages focus on:

• Token efficiency - Pre-processed, rich context
• LLM-first design - Built for AI consumption
• Professional quality - Production-ready, tested, documented

License

MIT License - see LICENSE for details.

Author

Om Surushe

• GitHub: @om-surushe
• LinkedIn: om-surushe
• npm: @om-surushe

---

Made with ❤️ and Bun for AI assistants