mcp-web-search
Provides DuckDuckGo web search and site fetching tools for Claude Desktop, enabling up to 10 search results and text content extraction from URLs.
README
mcp-web-search
A local MCP server that gives Claude web search and page-fetching capabilities — built to work around the network restrictions that Claude organisation accounts operate under.
Why this exists
Claude.ai organisation accounts restrict outbound network access to a fixed domain allowlist. General web search and browsing are blocked and regular members cannot change that. Personal accounts have built-in web search and don't have this problem.
This server runs locally on your machine — where the internet is unrestricted — and proxies web requests back to Claude via the MCP protocol. If you've ever hit host_not_allowed with no way to fix it, this is for you.
What it does
Two tools, nothing more:
| Tool | Description |
|---|---|
search_list |
Search the web via DuckDuckGo and return up to 10 ranked results |
access_site |
Fetch any URL, follow redirects, and return its content as plain text |
Tech stack
| Component | Technology | Version |
|---|---|---|
| Language | Python | ≥ 3.12 |
| MCP framework | FastMCP (mcp package) |
≥ 1.28.1 |
| Search backend | DuckDuckGo (ddgs package) |
≥ 9.14.4 |
| Package manager | uv | any recent |
Architecture
graph TD
Claude[Claude Desktop] -->|MCP stdio / HTTP| Server[mcp-web-search<br/>FastMCP server]
Server -->|HTTPS| DDG[DuckDuckGo API]
Server -->|HTTPS| Web[Any public URL]
Claude calls one of the two tools over MCP. The server makes the outbound request from the local machine (bypassing Claude's network restrictions) and returns structured JSON.
Key design decisions
No authentication on the server
Decision: The server accepts all requests without any auth token or API key.
Why: It only binds to 127.0.0.1 by default and is intended as a personal local tool — adding auth would be friction with no real security benefit in that context.
Do not change because: adding auth here would also require changes to every Claude Desktop config that uses this server.
Raw urllib instead of httpx / requests
Decision: access_site uses the stdlib urllib.request only.
Why: Avoids an extra dependency for a single HTTP call. ddgs already handles the search transport.
Do not change because: it keeps the dependency surface minimal; only revisit if redirect handling or TLS behaviour becomes a problem.
Setup & contributing
See CONTRIBUTING.md.
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。