Add web search to Claude Desktop with a self-hosted MCP server

In short: add Siftdog to Claude Desktop’s claude_desktop_config.json as a local stdio server. Claude gets four read-only tools; set SEARXNG_URL if you want web search, while extraction, crawling, and mapping work without it.

What Claude gets

Siftdog exposes four read-only MCP tools:

Search runs through your SearXNG instance. Siftdog does not require a search API key.

Add Siftdog to Claude Desktop

1. Install uv so Claude Desktop can run uvx.
2. Open Claude Desktop’s claude_desktop_config.json and add this entry under mcpServers:

{
  "mcpServers": {
    "siftdog": {
      "command": "uvx",
      "args": ["siftdog", "mcp"],
      "env": {
        "SEARXNG_URL": "http://your-searxng:8080"
      }
    }
  }
}

Replace http://your-searxng:8080 with the address of a SearXNG instance reachable from your computer. If you only need extraction, crawling, or mapping, you can remove the SEARXNG_URL entry. The search tool will need it when you want web results.

Save the file and restart Claude Desktop. Then ask Claude to use siftdog_search for a web query or siftdog_extract to read a URL.

Keep the setup local

This configuration starts Siftdog as a local process using stdio; it does not point Claude Desktop at a hosted Siftdog service. You can review the Siftdog repository for the full setup, including how to run SearXNG and Siftdog together with Docker Compose.

If you’re connecting to a Siftdog server that is already running, use its /mcp endpoint instead of this stdio configuration. For the equivalent local setup in Claude Code, see Add self-hosted web search to Claude Code with MCP.

Limits

The search tool depends on SearXNG and the public search engines it queries, so upstream rate limits can affect results. Siftdog extracts static HTML; it does not render JavaScript pages. If you need JavaScript rendering or a broader scraping platform, compare alternatives such as Firecrawl or crw.