Mcp Webgate

Gives Claude three tools to pull web content without blowing up your context window. webgate_query runs a search, fetches pages in parallel, strips out HTML junk with lxml, and optionally routes results through a local LLM to produce a compact Markdown summary with citations. webgate_fetch handles single URLs. Both tools hard-cap output size and apply BM25 reranking so your AI gets clean, ranked text instead of raw HTML soup. Works with SearXNG, Brave, Tavily, Exa, or SerpAPI as the search backend. The real win is turning 1.3 million tokens of raw search results into 1,450 tokens of structured content. Reach for this when you need web search that respects token budgets.