ihor-sokoliuk/mcp-searxng

What it solves

mcp-searxng provides a privacy-respecting way for AI assistants to perform web searches and read URL content. It eliminates the need for paid search-vendor API keys by allowing users to connect their AI tools to a self-hosted or trusted SearXNG instance.

How it works

It acts as a standalone Model Context Protocol (MCP) server written in Node.js. The server connects to one or more SearXNG instances via their HTTP JSON API, translating the AI assistant's requests into search queries and processing the results. It can be configured to use multiple replicas for failover or parallel search (fan-out).

Who it’s for

Users of MCP-compatible AI assistants (such as Claude Desktop, Cursor, and Windsurf) who want web search capabilities without relying on third-party proprietary search APIs, or those who prioritize privacy and control over their search data.

Highlights

• Privacy-focused search: Integrates with SearXNG to avoid third-party search operator trust.
• Advanced URL reading: Converts HTML to Markdown, extracts text from PDFs (up to 500 pages), and handles JSON/plain text.
• Browser solver integration: Supports FlareSolverr and Byparr to bypass challenges on uncached URLs.
• Instance failover: Supports multiple interchangeable SearXNG replicas with automatic failover or parallel merging of results.
• Intelligent caching: Uses in-memory caching with LFU eviction to reduce redundant requests.
• SSRF protection: Blocks private and internal URLs and redirects by default to enhance security.
• Flexible deployment: Supports both STDIO and HTTP transport modes, with Docker and NPM installation options.