WebShift is a Rust-based library and MCP server designed for processing web content by transforming noisy HTML into clean, structured text suitable for large language models (LLMs). It effectively removes unwanted elements such as scripts, ads, and navigation menus, ensuring that only relevant content is provided to the LLM, thus optimizing token usage. The tool offers various features including an HTML denoiser, text rewriter, web content client, and MCP server functionalities, all of which can be customized based on specific project needs.

With its focus on size management, WebShift allows users to enforce strict output limits, making it ideal for applications that require precise control over text size. The library supports multiple search backends and includes advanced features like query expansion and summarization for enhanced results. Though it is currently in beta, the core functionalities are stable, and the project is open to community feedback and contributions.