Your SearXNG instance, now wired for LLMs

What it does

mcp-searxng is a standalone Node.js process that acts as an MCP server between AI assistants and a SearXNG instance. It exposes two tools to clients like Claude or Cursor: web search with pagination, time filtering, and language selection, plus URL content extraction with section targeting and heading extraction. The assistant speaks MCP to this process; the process forwards queries to your SearXNG instance via its HTTP JSON API.

The interesting bit

This is fundamentally glue code: it neither installs into SearXNG nor bundles a search engine. The payoff is privacy—your AI assistant queries your own self-hosted instance instead of a commercial API—plus a caching layer for URL content that avoids redundant fetches.

Key highlights

• Self-hosted by design: you supply the SearXNG instance, keeping queries off third-party search APIs
• Two-tool surface: searxng_web_search for queries, web_url_read for content extraction with paragraph ranges and heading extraction
• Content caching with configurable TTL for URL reads
• Supports both STDIO and HTTP transports for the MCP connection
• Time range, language, and safe-search filtering built in

Caveats

• Requires a running SearXNG instance with JSON output explicitly enabled; a missing json format in settings.yml yields opaque 403 errors
• The project is a protocol translator, not a search engine—if you don’t already run SearXNG, this adds infrastructure overhead

Verdict

Worth a look if you already self-host SearXNG and want to give Cursor or Claude private web search without a third-party API middleman. Skip it if you were hoping for a batteries-included search backend.