Searxng Deepdive

Wraps SearXNG's meta-search API with four distinct tools: broad search across all engines, targeted search on specific engines like arxiv or pubmed, category-scoped search, and a lightweight HTML-to-Markdown URL reader with TOC extraction. The server dynamically injects your instance's actual engine and category lists into tool descriptions so the LLM knows what's available without guessing names. Multi-page fanout lets you request several result pages in one call with automatic deduplication. Validation catches common mistakes like passing an engine name to the category parameter. The URL reader handles static HTML, JSON, YAML, and TOML without a headless browser, with extraction modes for reading just headings or specific sections to save tokens on long documentation pages.