A Rust-based MCP server that turns messy web search results into clean, token-budgeted text ready for LLM consumption. It exposes three tools: webshift_query for full search pipelines (search, fetch, clean, BM25 rerank, optional summarization), webshift_fetch for single-page cleaning, and webshift_onboarding for agent configuration guidance. Supports eight search backends including SearXNG, Brave, Tavily, and Google, with hard caps on download size, per-page character limits, and total result budgets. Strips HTML noise (scripts, ads, nav elements), sterilizes Unicode and BiDi text, and enforces strict output sizes so you don't flood context windows. Ships as a single static binary with zero runtime dependencies. Reach for this when your agent needs web search that respects token budgets and delivers clean text instead of raw HTML soup.