🔍 Search Fusion MCP Server

🌏 中文文档

A High-Availability Multi-Engine Search Aggregation MCP Server providing intelligent failover, unified API, and LLM-optimized content processing. Search Fusion integrates multiple search engines with smart priority-based routing and automatic failover mechanisms.

🆕 What's New in v3.0.0: Major concurrency upgrade! Enhanced multi-threading support with thread-safe operations, intelligent connection pooling, and semaphore-based request limiting. Now supports 50+ concurrent searches without race conditions or data corruption!

✨ Features

🔄 Multi-Engine Integration

• Google Search - Premium performance with API key
• Serper Search - Google search alternative with advanced features
• Jina AI Search - AI-powered search with intelligent content processing
• DuckDuckGo - Free search, no API key required
• Exa Search - AI-powered semantic search
• Bing Search - Microsoft search API
• Baidu Search - Chinese search engine

🚀 Advanced Features

• Intelligent Failover - Automatic engine switching on failures or rate limits
• Priority-Based Routing - Smart engine selection based on availability and performance
• Unified Response Format - Consistent JSON structure across all engines
• Rate Limiting Protection - Built-in cooldown mechanisms
• 🔄 High Concurrency Support - Thread-safe operations with connection pooling
• ⚡ Performance Optimization - Async operations with semaphore-based concurrency control
• LLM-Optimized Content - Advanced web content fetching with pagination support
• Wikipedia Integration - Dedicated Wikipedia search tool
• Wayback Machine - Historical webpage archive search
• Environment Variable Configuration - Pure MCP configuration without config files
• 🌐 Enhanced Proxy Auto-Detection - Intelligent proxy detection with zero configuration

📊 Monitoring & Analytics

• Real-time engine status monitoring
• Success rate tracking
• Error handling and recovery
• Performance metrics

⚡ Concurrency & Performance

• Thread-Safe Operations - All engine statistics and state updates are protected by async locks
• Connection Pooling - Shared HTTP client with configurable connection limits (max 100 connections)
• Semaphore Control - Concurrent request limiting (max 30 simultaneous searches)
• Timeout Protection - 60-second search timeout prevents request accumulation
• Resource Management - Efficient memory usage with automatic connection cleanup
• Race Condition Prevention - Double-checked locking for SearchManager initialization

🏗️ Architecture

Search Fusion MCP Server
├── 🔧 Configuration Manager     # MCP environment variable handling
├── 🔍 Search Manager           # Multi-engine orchestration with concurrency control
├── ⚡ Concurrency Layer        # Thread-safe operations & performance optimization
│   ├── AsyncLock Protection    # Thread-safe state updates
│   ├── HTTP Connection Pool    # Shared client with connection limits
│   ├── Semaphore Control      # Concurrent request limiting (max 30)
│   └── Timeout Management     # 60s timeout protection
├── 🚀 Engine Implementations   # Individual search engines
│   ├── GoogleSearch            # Google Custom Search
│   ├── SerperSearch           # Serper API
│   ├── JinaSearch             # Jina AI Search
│   ├── DuckDuckGoSearch       # DuckDuckGo
│   ├── ExaSearch              # Exa AI
│   ├── BingSearch             # Bing API
│   └── BaiduSearch            # Baidu API
├── 🛠️ Advanced Fetcher         # Multi-method web scraping
└── 📡 MCP Server              # FastMCP integration

🚀 Quick Start

Installation

Option 1: Install from PyPI (Recommended)

pip install search-fusion-mcp

Option 2: Install from Source

git clone https://github.com/sailaoda/search-fusion-mcp.git
cd search-fusion-mcp
pip install -e .

🌐 Enhanced Proxy Auto-Detection (New in v2.0!)

Search Fusion now features intelligent proxy auto-detection inspired by concurrent-browser-mcp, providing seamless proxy support with zero configuration!

✨ Three-Layer Detection Strategy

1. Environment Variables - Highest priority, checks HTTP_PROXY, HTTPS_PROXY, ALL_PROXY
2. Port Scanning - Scans common proxy ports using socket connection testing
3. System Proxy - Detects OS-level proxy settings (macOS supported)

🔍 Supported Proxy Ports (Priority Order)

• 7890 - Clash default port
• 1087 - V2Ray common port
• 8080 - Generic HTTP proxy port
• 3128 - Squid proxy default port
• 8888 - Other proxy software port
• 10809 - V2Ray SOCKS port
• 20171 - Additional proxy port

🚀 Zero Configuration Usage

Just run directly - proxy will be auto-detected:

search-fusion-mcp

Manual override (if needed):

env HTTP_PROXY="http://your-proxy:port" search-fusion-mcp

📊 Detection Process

🔍 Checking environment variables...
🔍 Scanning proxy ports: [7890, 1087, 8080, ...]
✅ Local proxy port detected: 7890
🌐 Auto-detected proxy: http://127.0.0.1:7890

🆚 Comparison with concurrent-browser-mcp

MCP Integration

Environment Variable Configuration

Search Fusion uses pure MCP environment variable configuration without requiring config files.

MCP Client Configuration (PyPI Installation):

{
  "mcp": {
    "mcpServers": {
      "search-fusion": {
        "command": "search-fusion-mcp",
        "env": {
          "GOOGLE_API_KEY": "your_google_api_key",
          "GOOGLE_CSE_ID": "your_google_cse_id",
          "SERPER_API_KEY": "your_serper_api_key",
          "JINA_API_KEY": "your_jina_api_key",
          "EXA_API_KEY": "your_exa_api_key",
          "BING_API_KEY": "your_bing_api_key",
          "BAIDU_API_KEY": "your_baidu_api_key",
          "BAIDU_SECRET_KEY": "your_baidu_secret_key"
        }
      }
    }
  }
}

MCP Client Configuration (Source Installation):

{
  "mcp": {
    "mcpServers": {
      "search-fusion": {
        "command": "python",
        "args": ["-m", "src.main"],
        "cwd": "/path/to/your/search-fusion-mcp",
        "env": {
          "GOOGLE_API_KEY": "your_google_api_key",
          "GOOGLE_CSE_ID": "your_google_cse_id",
          "SERPER_API_KEY": "your_serper_api_key",
          "JINA_API_KEY": "your_jina_api_key",
          "EXA_API_KEY": "your_exa_api_key",
          "BING_API_KEY": "your_bing_api_key",
          "BAIDU_API_KEY": "your_baidu_api_key",
          "BAIDU_SECRET_KEY": "your_baidu_secret_key"
        }
      }
    }
  }
}

Supported Environment Variables

Alternative Variable Names:

# Google
GOOGLE_SEARCH_API_KEY    # Alternative to GOOGLE_API_KEY
GOOGLE_SEARCH_CSE_ID     # Alternative to GOOGLE_CSE_ID

# Serper
SERPER_SEARCH_API_KEY    # Alternative to SERPER_API_KEY

# Others follow similar pattern...

Engine Priority

Search engines are prioritized automatically:

1. Google Search (Priority 1) - Premium performance with API key
2. Serper Search (Priority 1) - Google alternative with advanced features
3. Jina AI Search (Priority 1.5) - AI-powered search with optional API key for advanced features
4. DuckDuckGo (Priority 2) - Free, no API key required
5. Exa Search (Priority 2) - AI-powered search with API key
6. Bing Search (Priority 3) - Microsoft search API
7. Baidu Search (Priority 3) - Chinese search engine

🛠️ MCP Tools

1. search

Perform web searches with intelligent engine selection and failover.

Parameters:

• query (required): Search query terms
• num_results (default: 10): Number of results to return
• engine (default: "auto"): Engine preference
  • "auto": Automatic engine selection (recommended)
  • "google": Prefer Google Search
  • "serper": Prefer Serper Search
  • "jina": Prefer Jina AI Search
  • "duckduckgo": Prefer DuckDuckGo
  • "exa": Prefer Exa Search
  • "bing": Prefer Bing Search
  • "baidu": Prefer Baidu Search

2. fetch_url

Fetch and process web content with intelligent pagination and multi-method fallback.

Parameters:

• url (required): Web URL to fetch
• use_jina (default: true): Whether to prioritize Jina Reader for LLM-optimized content
• with_image_alt (default: false): Whether to generate alt text for images
• max_length (default: 50000): Maximum content length per page (auto-paginate if exceeded)
• page_number (default: 1): Retrieve specific page from previously fetched content

Features:

• Intelligent Multi-Method Fallback: Tries Jina Reader → Serper Scrape → Direct HTTP
• Automatic Pagination: Splits large content into manageable pages
• Concurrent-Safe Caching: Unique page IDs prevent conflicts in high-concurrency scenarios
• LLM-Optimized Content: Clean markdown format optimized for AI processing

3. get_available_engines

Get current status and availability of all search engines.

4. search_wikipedia

Search Wikipedia articles for entities, people, places, concepts, etc.

Parameters:

• entity (required): Entity to search for
• first_sentences (default: 10): Number of sentences to return (0 for full content)

5. search_archived_webpage

Search archived versions of websites using Wayback Machine.

Parameters:

• url (required): Website URL to search
• year (optional): Target year
• month (optional): Target month
• day (optional): Target day

📖 API Examples

Basic Search

# Automatic engine selection
result = await search("artificial intelligence trends 2024")

# Prefer specific engine
result = await search("machine learning", engine="google")

Advanced Web Fetching

# Fetch with intelligent pagination
result = await fetch_url("https://example.com/long-article")

# If content is paginated, get additional pages
if result.get("is_paginated"):
    page_2 = await get_page(result["page_id"], 2)

Wikipedia Search

# Get Wikipedia summary
result = await search_wikipedia("Python programming language")

# Get full article
result = await search_wikipedia("Quantum computing", first_sentences=0)

🧪 Development

Development Setup

git clone https://github.com/sailaoda/search-fusion-mcp.git
cd search-fusion-mcp
pip install -r requirements.txt
pip install -e .

🔧 Configuration Guide

For detailed configuration instructions, see MCP_CONFIG_GUIDE.md.

📊 Performance

• Latency: Sub-second response times with caching
• Availability: 99.9% uptime with intelligent failover
• Throughput: Handles concurrent requests efficiently
• Scalability: Efficient resource utilization and concurrent processing

📈 Concurrency Benchmarks

Tested Performance (v3.0.0+):

• ✅ 50+ concurrent searches - No race conditions or data corruption
• ✅ Thread-safe statistics - Accurate request counting and error tracking
• ⚡ Connection pooling - Efficient HTTP resource management
• 🛡️ Timeout protection - 60s per request prevents system overload
• 📊 Real-time monitoring - Live engine status during high load

Recommended Limits:

• Concurrent searches: 10 (configurable via semaphore)
• Connection pool: 100 max connections, 20 keep-alive
• Request timeout: 60 seconds
• Memory usage: ~50MB baseline + ~2MB per concurrent request

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests for new functionality
5. Submit a pull request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🚨 Rate Limiting & Best Practices

• Google Search: 100 queries/day (free tier)
• Serper API: Varies by plan
• Jina AI: Rate limits apply based on subscription
• DuckDuckGo: No official limits, but use responsibly
• Other engines: Check respective API documentation

Always implement appropriate delays and respect rate limits to ensure sustainable usage.

📞 Support

• 📖 Documentation
• 🐛 Issue Tracker
• 💬 Discussions

Made with ❤️ for the MCP community