ihor-sokoliuk/mcp-searxng

What it solves

AI assistants like Claude or Cursor often lack real-time access to the internet or private search engines. This project bridges that gap by connecting these assistants to a SearXNG instance, allowing them to perform web searches and read website content without relying on proprietary or paid search APIs.

How it works

This is a standalone Node.js process that acts as a Model Context Protocol (MCP) server. It connects to a SearXNG instance (or multiple replicas) via its HTTP JSON API. When an AI assistant requests a search or a URL, this server queries SearXNG, handles pagination, manages failover between multiple instances, and can even convert webpage content into Markdown for the assistant to read.

Who it’s for

• AI Developers building agentic workflows that require real-time web data.
• Power Users of MCP-compatible clients (like Claude Desktop or Cursor) who want to provide their assistants with private, self-hosted, or free web search capabilities.
• Privacy-Conscious Users who want to use SearXNG to avoid sending all search queries directly to large commercial search engines.

Highlights

• Web Search & Reading: Supports general, news, and article queries, plus content-type-aware Markdown conversion for URLs.
• High Availability: Supports multiple SearXNG replicas with automatic failover or parallel fan-out to merge results.
• Intelligent Caching: Uses in-memory caching for both search results and URL content to reduce redundant network requests.
• Robustness: Includes HTML fallback for instances that disable JSON, SSRF protection for URL reading, and proxy support.