free-web-search-ultimate
Universal Search-First Knowledge Acquisition Plugin for LLMs. Enables real-time web search and deep page browsing via MCP or CLI. Zero-cost, privacy-first, supports DuckDuckGo, Bing, Google, Brave, Wikipedia, Arxiv, YouTube, Reddit and more.
README
<div align="center"> <img src="assets/banner.png" alt="Free Web Search Ultimate" width="100%"/> <h1>🔍 Free Web Search Ultimate</h1> <p><strong>Universal Search-First Knowledge Acquisition Plugin for LLMs</strong></p> <p><em>One install. Every LLM. Real-time knowledge. Zero cost.</em></p>
<br/>
Stop letting your LLM guess. Give it the ability to search.
By default, every LLM—Claude, GPT-4, Gemini—answers questions from training data that can be months or years out of date. This plugin introduces a Search-First Paradigm: the LLM is instructed to use real-time web search as its primary knowledge source, not a fallback.
🏗️ Architecture
<div align="center"> <img src="assets/architecture.png" alt="Architecture Diagram" width="90%"/> </div>
One plugin, every ecosystem. Whether you use Claude Desktop, Cursor, OpenClaw, or a custom LangChain agent, this plugin connects your LLM to the live web through a unified interface.
The free_web_search/skills/ directory contains the auto-discoverable SKILL.md that CLI-Anything compatible frameworks (such as OpenClaw) load automatically at startup — no manual configuration required.
🌟 The "Search-First" Paradigm
| Old Paradigm (Default LLM) | New Paradigm (This Plugin) |
|---|---|
| Answers from training data | Answers from live web search |
| Knowledge cutoff date | Always up-to-date |
| May hallucinate facts | Cites verifiable sources |
| Single knowledge source | Multi-source cross-validation |
When this plugin is installed, the AI agent is instructed to:
- Never Guess Facts — Use
search-webbefore answering any factual, technical, or real-time question. - Override Internal Knowledge — Even if the LLM "knows" the answer, it verifies via search for topics prone to change.
- Deep Verification — If search snippets are insufficient, use
browse-pageto read the full source. - Cite Sources — Always provide the source URLs in the final answer.
📦 Installation
pip install free-web-search-ultimate
Requirements: Python 3.10+
Or install from source:
git clone https://github.com/wd041216-bit/free-web-search-ultimate.git
cd free-web-search-ultimate
pip install -e .
🔌 Integration Guide
Claude Desktop & Cursor (via MCP)
Add to your claude_desktop_config.json or Cursor MCP settings:
{
"mcpServers": {
"free-web-search": {
"command": "free-web-search-mcp",
"args": []
}
}
}
That's it. Claude and Cursor will now have access to search_web and browse_page tools.
OpenClaw (via CLI-Anything)
# Install — the skill is auto-discovered from the bundled SKILL.md
pip install free-web-search-ultimate
OpenClaw reads the bundled SKILL.md and automatically loads the Search-First behavioral instructions.
LangChain / Custom Agents
from langchain.tools import Tool
import subprocess, json
def search_web(query: str) -> str:
result = subprocess.run(
["search-web", query, "--json"],
capture_output=True, text=True
)
data = json.loads(result.stdout)
return data.get("answer", "No results found.")
search_tool = Tool(
name="search_web",
func=search_web,
description="Search the web for real-time information. Use this before answering any factual question."
)
OpenAI Function Calling
tools = [
{
"type": "function",
"function": {
"name": "search_web",
"description": "Search the web for real-time information, news, or facts. Always call this before answering factual questions.",
"parameters": {
"type": "object",
"properties": {
"query": {"type": "string", "description": "The search query"},
"type": {"type": "string", "enum": ["text", "news", "images", "videos", "books"], "default": "text"}
},
"required": ["query"]
}
}
}
]
💻 CLI Usage
search-web — Web Search
# General knowledge
search-web "Python 3.13 new features"
# Breaking news
search-web "OpenAI GPT-5" --type news --timelimit w
# Images
search-web "neural network diagram" --type images
# Chinese search
search-web "人工智能最新进展" --region zh-cn
# JSON output for programmatic use
search-web "quantum computing" --json
browse-page — Deep Page Reading
# Read full page content
browse-page "https://docs.python.org/3/whatsnew/3.13.html"
# JSON output
browse-page "https://arxiv.org/abs/2303.08774" --json
🏆 Why This Over Alternatives?
| Feature | This Plugin | Tavily API | Serper API | Bing Search API |
|---|---|---|---|---|
| Cost | Free | $0.01/req | $0.001/req | $3/1000 req |
| API Key Required | No | Yes | Yes | Yes |
| Privacy | Local | Cloud | Cloud | Cloud |
| MCP Support | Yes | Partial | No | No |
| CLI-Anything | Yes | No | No | No |
| Image Search | Yes | No | Yes | Yes |
| Book Search | Yes | No | No | No |
| Browse Page | Yes | Yes | No | No |
📄 License
MIT License — free for personal and commercial use.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。