mcp-search-server 0.1.5

MCP server for web search, PDF parsing, and content extraction

MCP Search Server

mcp-name: io.github.KazKozDev/search

MCP (Model Context Protocol) server for web search, content extraction, and PDF parsing.

All tools work out of the box using free public APIs. No API keys required. No registration needed.

Context-Aware AI: Built-in tools for real-time datetime and geolocation detection give LLMs the ability to understand "here and now" - enabling timezone-aware responses, location-based content, and time-sensitive information without manual configuration.

Features

• DateTime Tool: Get current date and time with timezone awareness
• Geolocation: IP-based location detection with timezone, coordinates, and ISP info
• Web Search: Smart multi-engine search with automatic fallback
  • DuckDuckGo (primary): Fast, reliable, works out of the box
  • Brave Search (fallback): Browser-based with anti-bot bypass
  • Startpage (fallback): Privacy-focused Google proxy
  • Qwant (fallback): European search engine
• Wikipedia Search: Search and retrieve Wikipedia articles
• Web Content Extraction: Extract clean text from web pages using multiple parsing methods
• PDF Parsing: Extract text from PDF files
• Multi-Source Search: Parallel search across multiple sources
• Academic Search: Search arXiv, PubMed for scientific papers
• GitHub Search: Find repositories and README files
• Reddit Search: Search posts and comments
• News Search: GDELT global news database
• 🆕 Credibility Assessment: Bayesian source credibility scoring with 30+ signals, domain age (WHOIS), citation network (PageRank), and uncertainty quantification - no API keys required
• 🆕 Text Summarization: Multi-strategy summarization (TF-IDF extractive, keyword-based, heuristic) - fast, accurate, no API keys required

Installation

Prerequisites

• Python 3.10 or higher
• pip

Install from PyPI (recommended)

pip install mcp-search-server

Install from source

git clone https://github.com/KazKozDev/mcp-search-server.git
cd mcp-search-server
pip install -e .

Optional: Browser-based search engines

To enable Brave Search and Startpage with anti-bot bypass (using Playwright):

# Install optional browser dependencies
pip install -e ".[browser]"

# Install Firefox browser (recommended - more stable on macOS)
playwright install firefox

# Alternative: Install Chromium browser
playwright install chromium

Note: DuckDuckGo works perfectly without Playwright. Browser support is only needed for Brave and Startpage fallback engines.

Usage

Running the server

The server can be run directly:

python -m mcp_search_server.server

Or using the installed script:

mcp-search-server

Configuration for Claude Desktop

Add this to your Claude Desktop configuration file:

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

{
  "mcpServers": {
    "search": {
      "command": "python",
      "args": [
        "-m",
        "mcp_search_server.server"
      ]
    }
  }
}

Or if you installed it as a package:

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

Configuration for other MCP clients

The server uses stdio transport, so it can be integrated with any MCP client that supports stdio.

Available Tools

1. search_web

Search the web with smart multi-engine fallback (DuckDuckGo → Qwant → Brave → Startpage).

Parameters:

• query (string, required): The search query
• limit (integer, optional): Maximum number of results (default: 10)
• mode (string, optional): Search mode - 'web' (default) or 'news'
• timelimit (string, optional): Filter by time - 'd' (past day), 'w' (past week), 'm' (past month), 'y' (past year), null (all time, default)
• engine (string, optional): Specific search engine - 'duckduckgo', 'brave', 'startpage', 'qwant' (default: auto-fallback)
• use_fallback (boolean, optional): Enable automatic fallback to other engines (default: true)
• no_cache (boolean, optional): Disable cache (default: false)

Examples:

Auto-fallback search (recommended):

{
  "query": "Python async programming",
  "limit": 5,
  "use_fallback": true
}

Search using specific engine:

{
  "query": "machine learning",
  "limit": 10,
  "engine": "brave",
  "use_fallback": false
}

Search for recent news (past day):

{
  "query": "latest AI developments",
  "limit": 10,
  "mode": "news",
  "timelimit": "d"
}

2. search_wikipedia

Search Wikipedia for articles.

Parameters:

• query (string, required): The search query
• limit (integer, optional): Maximum number of results (default: 5)

Example:

{
  "query": "Machine Learning",
  "limit": 3
}

3. get_wikipedia_summary

Get a summary of a specific Wikipedia article.

Parameters:

• title (string, required): The Wikipedia article title

Example:

{
  "title": "Artificial Intelligence"
}

4. extract_webpage_content

Extract clean text content from a web page.

Parameters:

• url (string, required): The URL to extract content from

Example:

{
  "url": "https://example.com/article"
}

Features:

• Multiple parsing methods (Readability, Newspaper3k, BeautifulSoup)
• Automatic fallback if one method fails
• Cleans boilerplate content (ads, navigation, etc.)

5. parse_pdf

Extract text from PDF files.

Parameters:

• url (string, required): The URL of the PDF file
• max_chars (integer, optional): Maximum characters to extract (default: 50000)

Example:

{
  "url": "https://example.com/document.pdf",
  "max_chars": 100000
}

Features:

• Supports PyPDF2 and pdfplumber
• Automatic library selection

6. search_multi

Search multiple sources in parallel (web + Wikipedia).

Parameters:

• query (string, required): The search query
• web_limit (integer, optional): Max web results (default: 5)
• wiki_limit (integer, optional): Max Wikipedia results (default: 3)

Example:

{
  "query": "Python programming",
  "web_limit": 5,
  "wiki_limit": 3
}

Features:

• Runs searches in parallel for faster results
• Combines results from multiple sources
• Returns structured output with clear source attribution

7. get_current_datetime

Get current date and time with timezone information. Essential for time-aware AI responses.

Parameters:

• timezone (string, optional): Timezone name (default: "UTC")
• include_details (boolean, optional): Include additional details (default: true)

Example:

{
  "timezone": "Europe/Moscow",
  "include_details": true
}

Returns:

• ISO datetime string
• Date and time components
• Day of week, week number
• Multiple formatted representations
• Unix timestamp

Features:

• Supports 596+ timezones worldwide
• Automatic timezone conversion
• Detailed formatting options
• Graceful error handling for invalid timezones

8. list_timezones

List available timezones by region.

Parameters:

• region (string, optional): Region filter - "all", "Europe", "America", "Asia", "Africa", "Australia" (default: "all")

Example:

{
  "region": "Europe"
}

Features:

• Lists all available timezone names
• Filter by continent/region
• Useful for discovering correct timezone names

9. get_location_by_ip

Get geolocation information based on IP address. Returns country, city, timezone, coordinates, ISP, and more.

Parameters:

• ip_address (string, optional): IP address to lookup (e.g., "8.8.8.8"). If not provided, detects the server's public IP location.

Example:

{
  "ip_address": "8.8.8.8"
}

Returns:

• IP address
• Country, region, city, ZIP code
• Timezone (can be used with get_current_datetime!)
• Latitude and longitude coordinates
• ISP and organization information
• AS number

Features:

• Free API, no API key required
• Automatic timezone detection for location-aware responses
• Works with both IPv4 and IPv6
• Graceful error handling for invalid/private IPs
• Perfect companion to datetime tool for automatic timezone detection

Use Cases:

• Auto-detect user's timezone for time-aware responses
• Location-based content customization
• Network diagnostics and IP analysis
• Geographic data for analytics

10. assess_source_credibility 🆕

Assess the credibility of web sources using advanced Bayesian analysis with 30+ signals.

Parameters:

• url (string, required): URL to assess
• title (string, optional): Document title
• content (string, optional): Full text content (improves accuracy)
• metadata (object, optional): Structured metadata (year, authors, citations, doi, is_peer_reviewed)

Example:

{
  "url": "https://arxiv.org/abs/2301.00234",
  "title": "Deep Learning for Medical Imaging",
  "metadata": {
    "year": 2023,
    "is_peer_reviewed": true,
    "citations": 42
  }
}

Returns:

• Credibility score (0-1)
• Confidence interval (e.g., 0.75 ± 0.08)
• Category (academic, news, code, forum, blog, government)
• PageRank score from citation network
• 30+ individual signal scores
• Recommendation (✓✓ Excellent / ✓ Good / ⚠ Caution / ✗ Limited)

Features:

• Real Domain Age: WHOIS-based domain registration date checking
• Citation Network: PageRank algorithm for link analysis
• Bayesian Inference: Prior probabilities + likelihood + posterior
• 30+ Signals: Domain reputation, content quality, metadata analysis
• Uncertainty Quantification: Confidence intervals based on evidence
• No API Keys Required: All analysis runs locally

Optional Enhancement: Install WHOIS support for real domain age checking:

pip install mcp-search-server[credibility]

Documentation: See docs/CREDIBILITY_ASSESSMENT.md for detailed usage, examples, and technical details.

11. summarize_text 🆕

Summarize long text using multiple strategies (TF-IDF, keyword-based, or heuristic).

Parameters:

• text (string, required): Text to summarize
• strategy (string, optional): "auto" (default), "extractive_tfidf", "extractive_keyword", "heuristic"
• compression_ratio (number, optional): Target compression 0.1-0.9 (default: 0.3 = 30%)

Example:

{
  "text": "Long article text here...",
  "strategy": "extractive_tfidf",
  "compression_ratio": 0.3
}

Returns:

• Summary text
• Method used (extractive-tfidf, extractive-keyword, heuristic-3sent)
• Statistics (original/summary length, compression ratio, sentences)

Strategies:

• extractive_tfidf (best): Uses TF-IDF scoring to select important sentences. Requires NLTK.
• extractive_keyword: Prioritizes sentences with entities and key terms. Requires NLTK.
• heuristic: Ultra-fast fallback (first + middle + last sentences). No dependencies.
• auto: Automatically picks best available strategy.

Features:

• Fast: ~50ms for typical article (with NLTK), ~5ms (heuristic)
• No API Keys: All processing local
• Smart Selection: Maintains original sentence order
• Graceful Degradation: Falls back if NLTK unavailable

Optional Enhancement: Install NLTK for better quality:

pip install mcp-search-server[summarizer]

Use Cases:

• Summarize web articles before credibility assessment
• Condense research papers for quick review
• Extract key points from long documents
• Generate previews for search results

Development

Install development dependencies

pip install -e ".[dev]"

Running tests

pytest

Code formatting

black src/

Linting

ruff check src/

Architecture

Tools

• DuckDuckGo Search (tools/duckduckgo.py)

  • Async web scraping from DuckDuckGo HTML and Lite versions
  • Result caching (24 hours)
  • Retry logic with backoff

• Wikipedia (tools/wikipedia.py)

  • Wikipedia API integration
  • Article search and summary retrieval
  • HTML cleaning

• Link Parser (tools/link_parser.py)

  • Multiple parsing methods (Readability, Newspaper3k, BeautifulSoup)
  • Early exit optimization
  • Content cleaning

• PDF Parser (tools/pdf_parser.py)

  • PyPDF2 and pdfplumber support
  • Automatic library selection
  • Page-by-page extraction with limits

Caching

The server uses local caching for search results:

• Location: ~/.mcp-search-cache/
• TTL: 24 hours
• Format: JSON

Troubleshooting

PDF parsing not working

Install one of the PDF libraries:

pip install PyPDF2
# or
pip install pdfplumber

Web content extraction fails

The server tries multiple methods automatically:

1. Readability (best for articles)
2. Newspaper3k (good for news sites)
3. BeautifulSoup (fallback for all sites)

If all methods fail, check:

• The URL is accessible
• The site doesn't block automated access
• Your internet connection

Wikipedia search returns no results

• Check your internet connection
• Try a different search term
• The Wikipedia API might be temporarily unavailable

License

MIT

Contributing

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