v0.2.1 - Fix CLI Entry Point

v0.2.1 - Hotfix: CLI Entry Point

Fixes the "Plugin process exited with code 1" error in LM Studio and other MCP clients.

🐛 Bug Fix

• Added missing run() function as CLI entry point for mcp-search-server command
• This fixes the issue where the server would crash on startup with exit code 1

📦 Installation

pip install --upgrade mcp-search-server

🔧 For LM Studio Users

After upgrading, configure in LM Studio settings:

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

Or simply:

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

Both should now work correctly!

---

Full Changelog: v0.2.0...v0.2.1

v0.2.0 - Modular Registry Architecture

Release v0.2.0: Modular Registry Architecture

Major refactoring introducing a flexible, extensible architecture with enhanced tool discovery capabilities.

🎯 Highlights

• 27 tools (up from 24)
• Modular registry architecture for dynamic tool loading
• Meta-tools for self-discovery and documentation
• 7 organized categories: meta, web, knowledge, social, analysis, context, files

✨ New Features

Meta-Tools for Tool Discovery

• search_tools: Find tools by keywords, categories, or use cases with relevance scoring
• list_tool_categories: Browse all available tool categories
• get_tool_info: Get detailed documentation for any tool

Architecture Improvements

• Dynamic tool loading via registry system
• Category-based organization
• Enhanced extensibility and maintainability
• Server code reduced from 800+ to 139 lines

🔄 Breaking Changes

• Tools reorganized into category subdirectories:

  • tools/web/ - Web search, content extraction, PDFs
  • tools/knowledge/ - Wikipedia, arXiv, PubMed, GDELT
  • tools/social/ - GitHub, Reddit
  • tools/analysis/ - Credibility, summarizer, calculator
  • tools/context/ - Date/time, geolocation
  • tools/files/ - File management
  • tools/meta/ - Tool discovery

• Renamed tool: search_web → search_duckduckgo

• New import paths for all tools (backward incompatible)

📦 Installation

pip install mcp-search-server==0.2.0

Or upgrade from previous version:

pip install --upgrade mcp-search-server

📚 Documentation

Updated README with:

• Complete list of all 27 tools
• Architecture section explaining the registry system
• Category-based tool organization
• Usage examples for meta-tools

🔧 Technical Details

New Components:

• registry/ - Tool registry, loader, and category manager
• tools/base.py - Base classes for all tools
• tools/meta/ - Meta-tools for introspection
• config/ - YAML configuration files

Code Quality:

• All tests passing (58 tests ✅)
• Black formatting applied
• Ruff linting passing
• Type hints throughout

Stats:

• 44 files changed
• 4,025 insertions
• 2,810 deletions
• Cleaner, more maintainable codebase

🙏 Credits

Built with Claude Code

---

Full Changelog: v0.1.9...v0.2.0

v0.1.9

What's Changed

Bug Fixes

• Fix outputSchema validation errors for all 24 MCP tools
• Fix assess_source_credibility tuple→list conversion for JSON schema
• Add ^ to ** power operator conversion in calculator
• Fix gdeltdoc package name in dependencies

Testing

• All 24/24 tools pass validation tests

Full Changelog: v0.1.8...v0.1.9

v0.1.8

Release v0.1.8

Fix: Structured Output for MCP Tools

All 24 tools with outputSchema now return proper structured content as tuples (text_content, structured_dict) instead of embedding JSON in text comments.

This fixes LangChain/LangGraph validation errors:

Output validation error: outputSchema defined but no structured output returned

Changed Tools

• search_web, search_maps, search_wikipedia, get_wikipedia_summary
• extract_webpage_content, parse_pdf, get_current_datetime, get_location_by_ip
• search_arxiv, search_github, get_github_readme, search_reddit
• get_reddit_comments, search_pubmed, search_gdelt, assess_source_credibility
• summarize_text, get_wikipedia_content, read_file, write_file
• append_file, list_files, delete_file, calculate

Upgrade

pip install --upgrade mcp-search-server

v0.1.7

Release v0.1.7

• Fix version sync with pyproject.toml

v0.1.6

Release v0.1.6

• Fix Windows path compatibility in tests
• Apply black formatting
• Fix failing tests: calculator and file_manager
• Fix linting errors
• Sync version in init.py

v0.1.5: Smart Multi-Engine Search with Playwright Support

Release Notes v0.1.5

Release Date: December 25, 2025

🎉 Major Features

Smart Multi-Engine Search with Automatic Fallback

Version 0.1.5 introduces a revolutionary smart search system with automatic fallback across multiple search engines. Never worry about search failures again!

How it works:

1. DuckDuckGo (primary) - Fast, reliable, works out of the box
2. Qwant (backup) - European search engine
3. Brave Search (fallback) - Browser-based with anti-bot bypass
4. Startpage (fallback) - Privacy-focused Google proxy

The system automatically falls back to the next engine if the previous one returns less than 3 results.

Playwright Browser Automation

New optional browser automation support for JavaScript-heavy sites:

• Firefox (primary) - More stable on macOS
• Chromium (fallback) - Alternative browser
• Anti-bot bypass - Defeats Cloudflare, JavaScript challenges
• Smart rendering - Only used when needed

🚀 New Features

1. Search Engine Selection

Choose your preferred search engine:

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

Supported engines: duckduckgo, qwant, brave, startpage

2. Fallback Control

Enable or disable automatic fallback:

{
  "query": "Python programming",
  "use_fallback": true
}

3. Browser Dependencies (Optional)

Install browser support separately:

pip install -e ".[browser]"
playwright install firefox

4. Enhanced Documentation

• MIGRATION.md - Complete migration guide from v0.1.4
• FALLBACK_ENGINES.md - Detailed engine documentation
• requirements-browser.txt - Browser-specific dependencies

📊 Performance

Engine	Speed	Reliability	Browser Required
DuckDuckGo	~1-2s	99%	❌ No
Qwant	~1-2s	70%	❌ No
Brave	~3-5s	95%	✅ Yes
Startpage	~3-5s	80%	✅ Yes

🔧 Breaking Changes

None - This is a fully backward-compatible release. Your existing code will continue to work without any changes.

📝 Migration Guide

Option 1: No Changes (Recommended)

Your existing code works as-is:

{
  "query": "Python programming",
  "limit": 10
}

Option 2: Enable Browser Support (Optional)

For maximum reliability with Brave/Startpage:

pip install playwright
playwright install firefox

Option 3: Use Specific Engine

Choose your preferred search engine:

{
  "query": "AI news",
  "engine": "brave",
  "use_fallback": false
}

🐛 Bug Fixes

• Fixed: Chromium crashes on macOS (SEGV_ACCERR signal)
• Fixed: Anti-bot protection blocking Brave Search
• Fixed: Brave Search HTML parsing for new structure
• Fixed: Browser detection flags

🎯 Recommendations

For Production

• Use default settings (auto-fallback)
• Optionally install Playwright for maximum reliability

For Maximum Speed

• Disable fallback: "use_fallback": false
• Don't install Playwright

For Maximum Reliability

• Install Playwright with Firefox
• Use auto-fallback: "use_fallback": true

📚 Documentation

• Migration Guide
• Search Engine Documentation
• README
• Changelog

🙏 Credits

Special thanks to:

• Playwright team for excellent browser automation
• DuckDuckGo for reliable search API
• The open-source community

🔗 Links

• PyPI: https://pypi.org/project/mcp-search-server/
• GitHub: https://github.com/KazKozDev/mcp-search-server
• Issues: https://github.com/KazKozDev/mcp-search-server/issues

📦 Installation

# Basic installation (DuckDuckGo + Qwant)
pip install mcp-search-server==0.1.5

# With browser support (all engines)
pip install mcp-search-server[browser]==0.1.5
playwright install firefox

---

Full Changelog: https://github.com/KazKozDev/mcp-search-server/blob/main/CHANGELOG.md

🤖 Generated with Claude Code

v0.1.4 - Advanced Content Extraction & DuckDuckGo Improvements

🎉 What's New in v0.1.4

🚀 Major Features

• Advanced Article Extraction with Trafilatura integration
• Undetected ChromeDriver - bypass bot protection on Medium, Cloudflare, etc.
• Content Caching System - 24-hour cache with 1000-entry limit
• 6-Method Fallback Chain for maximum content extraction success
• Metadata Extraction - title, author, date, description, language

🔍 DuckDuckGo Search Improvements

• Removed region parameter for better language auto-detection
• Fixed Bulgarian results issue for Russian queries
• DuckDuckGo now auto-detects optimal region from query language
• Cleaner API without unused parameters

⚡ Performance Improvements

• Success rate improved from 70% to 83%
• Cache response time: <0.1s vs 2-5s for fresh requests
• Early exit optimization saves 60-70% processing time
• Support for bot-protected sites (Medium, Forbes, Russian news)

🛠️ Technical Enhancements

• 3 retry attempts with exponential backoff
• 5 different User-Agent rotation
• Rate limiting support (429 status)
• SSL security with certifi
• Better async compatibility
• Human-like scraping behavior

🐛 Bug Fixes

• Fixed async deprecation warnings
• Better encoding handling (UTF-8 → Latin-1 fallback)
• Improved BeautifulSoup content detection
• Newspaper3k blocking issues resolved

📦 Installation

pip install mcp-search-server

🔗 Links

• Full Changelog
• Documentation

v0.1.3 - Advanced Tools Integration

🎉 Major Release: Advanced Tools Integration

New Features

🔍 Credibility Assessment

• Bayesian credibility scoring with 30+ signals
• Domain age verification via WHOIS
• PageRank-based citation network analysis
• Uncertainty quantification with confidence intervals
• Category-specific priors (Academic, News, Code, Forum, etc.)
• No API keys required

📝 Text Summarization

• Multi-strategy summarization (TF-IDF, keyword-based, heuristic)
• Configurable compression ratio (10%-90%)
• Automatic strategy selection
• Graceful degradation without NLTK
• No API keys required

🔬 Academic & Scientific Search

• arXiv Search: Scientific papers and preprints
• PubMed Search: Medical and biomedical literature

🌐 Extended Search Capabilities

• GitHub Search: Find repositories and README content
• Reddit Search: Search posts and comments
• GDELT Search: Global news database access

Bug Fixes

• Fixed pytest collection issue with manual test script

Improvements

• Enhanced README with comprehensive feature documentation
• Added optional dependency groups for credibility and summarization
• Improved error handling across all tools
• Performance optimizations

Installation

Basic:

pip install mcp-search-server

With credibility assessment:

pip install mcp-search-server[credibility]

With text summarization:

pip install mcp-search-server[summarizer]

Full installation:

pip install mcp-search-server[credibility,summarizer]

Breaking Changes

None - fully backward compatible

Full Changelog: v0.1.2...v0.1.3

v0.1.2

Added

• Selenium-based DuckDuckGo search implementation
• • Example MCP client configuration file (mcp_config.json)

Changed

• DuckDuckGo search now uses Selenium by default

Removed

• DuckDuckGo images/videos helper module (media_search)
• duckduckgo-search dependency
• healthcheck tool