Add self-hosted web search to Claude Code with MCP

In short: one command gives Claude Code four web tools that run on your machine.

claude mcp add siftdog -- uvx siftdog mcp

What you get

All four are read-only.

Option 1: run it locally

You need uv. Extract, crawl and map work right away. Search needs a SearXNG instance, which you can start with Siftdog’s config:

git clone https://github.com/khsarvar/siftdog && cd siftdog
docker run -d -p 8080:8080 -e SEARXNG_SECRET=$(openssl rand -hex 16) \
  -v "$PWD/docker/searxng/settings.yml:/etc/searxng/settings.yml:ro" searxng/searxng

Then add Siftdog with the SearXNG address:

claude mcp add siftdog -e SEARXNG_URL=http://localhost:8080 -- uvx siftdog mcp

Option 2: connect to a running server

If you run the full stack with Docker Compose, connect over HTTP instead:

docker compose up -d
claude mcp add --transport http siftdog http://localhost:8000/mcp

If the server has API_KEYS set, add --header "Authorization: Bearer <key>".

Check it

claude mcp get siftdog

It should show ✔ Connected. Then ask Claude something like “Use siftdog to find what’s new in Python 3.14, with sources.”

Tips

• Large sites: map first, then extract or crawl only the paths you need.
• Context: page content is cut at 4,000 characters per page by default (max_chars).
• Rate limits: public search engines block heavy bursts. If every engine is blocked, the tool returns an error naming them instead of empty results.

Other MCP clients

Claude Desktop, Cursor and most clients take the same command as JSON:

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