Open WebSearch

What Aas-ee/open-webSearch MCP server does

The Aas-ee/open-webSearch MCP server exposes web search and content retrieval through MCP, a command-line interface, and an optional local HTTP daemon. Search requests can use engines including Bing, Baidu, DuckDuckGo, Brave, Exa, CSDN, Juejin, Startpage, Sogou, and Hacker News. Linux.do is listed as temporarily unsupported.

Search responses contain structured result data such as titles, URLs, and descriptions. The server can also fetch content from individual CSDN pages, GitHub README files, and generic HTTP(S) pages or Markdown content. For extraction, searching first and then fetching a specific result is recommended because some homepages and JavaScript-heavy pages may not provide readable text.

How it works

For MCP use, the package can run through an stdio-compatible entrypoint. It also supports HTTP mode or both HTTP and stdio, depending on the MODE setting. The default port for HTTP operation is 3000. A separate local daemon can be started with open-websearch serve; its documented endpoints include health checks, search, and fetch operations. The open-websearch status command checks the daemon state.

The command-line interface supports one-shot searches, while the daemon stays available for repeated requests. Search actions can use the local daemon automatically when it is available, or a specific daemon can be selected with --daemon-url. HTTP proxy support is available for restricted network environments. Installation proxy settings and runtime proxy settings are separate concerns.

Setup and configuration

Install and launch the MCP entrypoint with:

npx -y open-websearch@latest

The Aas-ee/open-webSearch MCP server does not require API keys or authentication. Optional settings include DEFAULT_SEARCH_ENGINE, ALLOWED_SEARCH_ENGINES, and SEARCH_MODE. SEARCH_MODE can select request-based or Playwright-based searching, or allow automatic selection. ENABLE_CORS and CORS_ORIGIN configure CORS behavior when HTTP access is enabled.

Set USE_PROXY=true and provide PROXY_URL to route runtime requests through an HTTP proxy. PORT changes the HTTP port, and MODE selects both, http, or stdio. FETCH_WEB_INSECURE_TLS can disable TLS verification for the web-content fetch request leg, but does not change Playwright browser navigation. This option should be limited to cases involving broken certificate chains.

Tools and capabilities

• Search across supported engines without an API key.
• Select a default engine and restrict the engines available to clients.
• Control the number of returned results.
• Fetch generic web or Markdown content.
• Retrieve GitHub README and CSDN page content.
• Run as an MCP server, one-shot CLI, or local HTTP daemon.

The project also documents an optional agent skill that helps discover and activate an available open-websearch path. That skill complements the MCP server, CLI, or daemon rather than replacing them.