web-search-mcp

Why we included this project

Most people running local LLMs hit the same wall: the model's knowledge stops at its training cutoff. This server plugs that gap by handing the model live web search through the Model Context Protocol, with no search API keys to sign up for. It keeps several engines in rotation, Bing, Brave, and DuckDuckGo, and falls back to a real browser when a search endpoint blocks a plain request, so one flaky engine rarely kills the whole lookup. Alongside full searches it can return just the result snippets or extract the main content of a single page, stripping out navigation and ads. If you use LM Studio or LibreChat, this is a self-contained way to make your model ground its answers in current pages.