MCP Web Search Server

A Model Context Protocol (MCP) server that provides multi-engine web search capabilities with intelligent content extraction using a hybrid approach.

Features

• 🔍 Hybrid Search Engine: Fast goquery-based search results + intelligent chromedp content extraction
• 🌐 Multi-Engine Support: Bing, Brave, and DuckDuckGo with smart fallback mechanisms
• 📄 Intelligent Content Extraction: Advanced article parsing with multiple content selectors
• 🚀 Concurrent Processing: Parallel content extraction with rate limiting
• 🤖 AI-Ready Summaries: Aggregated content optimized for AI analysis and summarization
• 🛠️ MCP Protocol: Full compliance with Model Context Protocol specification

Installation

Via go install

go install github.com/liliang-cn/mcp-websearch-server@latest

From Source

git clone https://github.com/liliang-cn/mcp-websearch-server
cd mcp-websearch-server
go build -o mcp-websearch-server

Usage

Standalone

# Show help
mcp-websearch-server --help

# Run the server (stdio mode)
mcp-websearch-server

Integration with Claude Desktop

Add to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "websearch": {
      "command": "mcp-websearch-server"
    }
  }
}

If installed via go install, make sure ~/go/bin is in your PATH.

Using with Claude Code (Skills)

This project includes a SKILL.md definition that can be used with Claude Code. This skill provides specialized instructions for Claude to effectively use the web search tools for deep research and fact-checking.

To use it:

1. Ensure the MCP server is configured in your claude_desktop_config.json (as above).
2. Copy SKILL.md to your skills directory:

   mkdir -p ~/.claude/skills
   cp SKILL.md ~/.claude/skills/web-search.md
   

3. You can now use the skill in Claude Code via /web-search or by asking natural language questions that trigger the skill.

Available Tools

🔍 websearch_basic

Basic web search returning titles, URLs and snippets from a single search engine using the hybrid approach.

Parameters:

• query (string, required): The search query
• max_results (int, optional): Maximum results to return (default: 10)

📄 websearch_with_content

Web search with intelligent content extraction from result pages using chromedp.

Parameters:

• query (string, required): The search query
• max_results (int, optional): Maximum results to return (default: 5)
• extract_content (bool, optional): Extract full page content (default: true)

🚀 websearch_multi_engine

Comprehensive search across multiple engines (Bing, Brave, DuckDuckGo) with content extraction.

Parameters:

• query (string, required): The search query
• max_results (int, optional): Maximum results to return (default: 3)
• engines (array, optional): Search engines to use ["bing", "brave", "duckduckgo"] (default: all)

🤖 websearch_ai_summary

Search and return AI-ready aggregated content optimized for analysis and summarization.

Parameters:

• query (string, required): The search query
• max_results (int, optional): Maximum results to return (default: 3)

Returns: Formatted markdown content with proper structure for AI processing.

Architecture

mcp-websearch-server/
├── main.go                     # Entry point with CLI flags
├── mcp/                        # MCP protocol implementation
│   └── server.go              # MCP server and tool registration
├── search/                     # Search engine implementations
│   ├── interface.go           # Common interfaces
│   ├── hybrid_searcher.go     # Hybrid multi-engine searcher
│   ├── multi_engine.go        # Basic multi-engine orchestration
│   ├── bing_goquery.go        # Fast Bing search with goquery
│   ├── brave_goquery.go       # Fast Brave search with goquery
│   ├── duckduckgo_goquery.go  # Fast DuckDuckGo search with goquery
│   ├── bing.go               # Original Bing search (chromedp)
│   ├── brave.go              # Original Brave search (chromedp)
│   └── duckduckgo.go         # Original DuckDuckGo search (chromedp)
├── extraction/                 # Content extraction
│   ├── hybrid_extractor.go   # Intelligent chromedp-based extraction
│   └── chromedp.go           # Basic browser-based extraction
├── examples/                   # Demo applications
│   ├── basic_search_demo/     # Basic search functionality demo
│   ├── hybrid_search_demo/    # Hybrid search with content extraction
│   └── mcp_tools_demo/        # MCP server tools demonstration
└── utils/                     # Utilities
    └── retry.go              # Retry logic with backoff

Hybrid Approach

The server uses a sophisticated hybrid approach for optimal performance:

1. Fast Search Results (goquery)

• Bing: Scrapes www.bing.com/search with proper CSS selectors
• Brave: Scrapes search.brave.com/search for results
• DuckDuckGo: Scrapes duckduckgo.com with lite interface
• Benefits: Fast response times, reliable result parsing

2. Intelligent Content Extraction (chromedp)

• Article Detection: Uses advanced selectors to find main content
• Content Cleaning: Removes scripts, styles, and navigation elements
• Fallback Strategy: Falls back to paragraph extraction if article content not found
• Benefits: High-quality content extraction, JavaScript handling

3. AI-Ready Aggregation

• Structured Output: Properly formatted markdown for AI processing
• Content Summarization: Truncates content intelligently at sentence boundaries
• Multi-Source: Combines content from multiple search engines
• Benefits: Optimized for AI analysis and summarization

Development

Prerequisites

• Go 1.21 or higher
• Chrome/Chromium browser (for content extraction)

Building

# Build the server
go build -o mcp-websearch-server

# Run tests
go test ./...

# Run tests with coverage
go test -cover ./...

# Format code
go fmt ./...

# Lint (requires golangci-lint)
golangci-lint run

Testing

The project includes comprehensive unit tests with 60%+ coverage:

# Run all tests
go test ./...

# Run with verbose output
go test -v ./...

# Generate coverage report
go test -coverprofile=coverage.out ./...
go tool cover -html=coverage.out

Example Applications

# Test basic search functionality
go run ./examples/basic_search_demo/main.go

# Test hybrid search with content extraction
go run ./examples/hybrid_search_demo/main.go

# Test MCP server tools
go run ./examples/mcp_tools_demo/main.go

How It Works

1. Search Request: Receives search query via MCP protocol
2. Engine Selection: Uses goquery-based engines for fast results
3. Search Execution: Performs HTTP-based search with proper headers
4. Content Extraction: Uses chromedp for intelligent content extraction
5. Aggregation: Combines and formats content for AI analysis
6. Response: Returns structured results via MCP protocol

Search Engine Priority

The hybrid searcher prioritizes engines in this order:

1. DuckDuckGo - Primary engine (privacy-focused)
2. Bing - First fallback (comprehensive results)
3. Brave - Second fallback (independent search)

If one engine fails, the server automatically tries the next available engine.

Error Handling

• Implements retry logic with exponential backoff
• Graceful fallback to alternative search engines
• Structured error messages via MCP protocol
• Timeout handling for long-running operations
• Rate limiting for content extraction

Performance

• Search Speed: ~200-500ms per search using goquery
• Content Extraction: ~2-5s per page using chromedp
• Concurrent Extraction: Limited to 2-3 simultaneous browser instances
• Memory Usage: Optimized with proper context cleanup

Dependencies

• MCP Go SDK: Model Context Protocol implementation
• chromedp: Browser automation for content extraction
• goquery: Fast HTML parsing and scraping
• Standard Library: HTTP client, context, sync primitives

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE file for details

Acknowledgments

• Built with MCP Go SDK
• Uses chromedp for browser automation
• Uses goquery for HTML parsing
• Implements Model Context Protocol specification