mcp-searxng

What it does

The server connects Claude or other MCP clients to any SearXNG instance, enabling private web search without relying on public search engines. It exposes two primary tools: searxng_web_search for general web queries with pagination, time filtering, and language selection, and web_url_read for extracting and converting web page content to markdown with optional section filtering and paragraph selection. Results are cached with TTL to reduce redundant requests.

Who it's for

Backend engineers and privacy-conscious users operating a self-hosted SearXNG instance who want their AI assistants to search the web without sending queries to centralized search providers. Developers integrating Claude or Cursor with private search infrastructure.

Common use cases

• Search the web with pagination and time-range filters (past day, month, or year)
• Extract markdown-formatted content from specific URLs or webpage sections
• Filter search results by language preference
• Cache URL content to reduce load on the SearXNG instance
• Build AI workflows that combine search and content extraction

Setup pitfalls

• SEARXNG_URL is mandatory — point it to your running SearXNG instance, not the GitHub repo
• If using Docker, environment variables must be passed via -e flags; they do not inherit automatically from the host
• Requires a working SearXNG instance on the network — the server will fail if the instance is unreachable
• The server reads and writes the filesystem for caching, so run it in an isolated environment if untrusted queries are expected