Searxng Docker

SearXNG Docker Stack

Self-hosted SearXNG metasearch with Redis caching, a Python query CLI, an MCP server for AI tools, and optional LLM-backed structured extraction. Clone, configure .env, run docker compose up -d, then search in the browser, from scripts, or through Cursor/Claude.

Stack

Quick start

Prerequisites

• Docker and Docker Compose
• For extract: an OpenRouter API key in .env

Configure and run

cp env.example .env
# Edit .env: set SEARXNG_SECRET_KEY (see below) and OPENROUTER_API_KEY if using extract
openssl rand -hex 32   # paste into SEARXNG_SECRET_KEY
docker compose up -d

Minimum .env values:

SEARXNG_PROTOCOL=http
SEARXNG_HOST=localhost
SEARXNG_PORT=7777
SEARXNG_BASE_URL=${SEARXNG_PROTOCOL}://${SEARXNG_HOST}:${SEARXNG_PORT}/
SEARXNG_SECRET_KEY=<from openssl rand -hex 32>
SEARXNG_MCP_PORT=7778

For structured extraction, keep EXTRACT_ENABLED=true (default in env.example) and set:

OPENROUTER_API_KEY=sk-or-...

To disable the extract tool (search/fetch/crawl only), set EXTRACT_ENABLED=false in .env.

Verify

docker compose ps
curl -sf "http://localhost:7777/search?q=test&format=json" | head -c 200
curl -sf http://localhost:7778/health

• Search UI: http://localhost:7777 (or your SEARXNG_PORT)
• MCP (streamable HTTP): http://localhost:7778/mcp
• REST mirror: http://localhost:7778

Usage

Pick the interface that fits your workflow. All paths use the same private SearXNG instance.

Web UI

Open the search URL in a browser. No extra setup beyond Quick start.

SearXNG search API

Direct JSON search against SearXNG (no MCP):

curl "http://localhost:7777/search?q=python+docker&format=json"
curl "http://localhost:7777/search?q=docker&categories=general,it&format=json"

Response fields, engines, and error handling: API_DOCUMENTATION.md.

Python: searxng_search.py

Command-line wrapper around the SearXNG API (stdlib only):

python searxng_search.py "your search term"
python searxng_search.py "docker" --categories general,it
python searxng_search.py "python" --output json

More flags and examples: QUERY_SCRIPT_README.md.

MCP server: tools and when to use them

The MCP server (searxng-mcp) exposes four capabilities. In agents, fetch returns full page text; extract fetches then returns JSON shaped by your schema and prompt—prefer extract on noisy pages when you want fields, not raw HTML.

REST examples (same host/port as health check):

# Search
curl -sS -X POST http://localhost:7778/search \
  -H "Content-Type: application/json" \
  -d '{"query": "python programming", "categories": "general,it"}'

# Fetch page text
curl -sS -X POST http://localhost:7778/fetch \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

# Crawl related pages
curl -sS -X POST http://localhost:7778/crawl \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com", "filters": ["documentation"], "subpage_limit": 5}'

# Extract structured JSON (EXTRACT_ENABLED=true, OPENROUTER_API_KEY set)
curl -sS -X POST http://localhost:7778/extract \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com","json_schema":{"type":"object","properties":{"title":{"type":"string"}},"required":["title"]}}'

Python: extract_url.py

CLI for /extract without hand-writing JSON Schema:

python extract_url.py "https://example.com/article" --keys title,author,summary

python extract_url.py "https://example.com/article" \
  --keys title,summary \
  --prompt "Main article only; ignore navigation and related links"

python extract_url.py "https://example.com/article" \
  --json '{"title":"Page title","summary":"Two-sentence summary"}'

Stdout is the extracted data object (indented JSON). Optional env: SEARXNG_MCP_HOST, SEARXNG_MCP_PORT (default localhost:7778), EXTRACT_TIMEOUT (default 120).

Cursor and Claude

1. Point the client at the MCP server (~/.cursor/mcp.json or Claude Desktop config).

Streamable HTTP (recommended when the container port is reachable):

{
  "mcpServers": {
    "searxng": {
      "url": "http://localhost:7778/mcp",
      "description": "Private SearXNG: search, fetch, crawl, extract"
    }
  }
}

stdio via Docker exec (no published port required):

{
  "mcpServers": {
    "searxng": {
      "command": "docker",
      "args": ["exec", "-i", "searxng-mcp", "python", "server.py"],
      "description": "Private SearXNG: search, fetch, crawl, extract"
    }
  }
}

Claude Desktop paths: macOS ~/Library/Application Support/Claude/claude_desktop_config.json, Windows %APPDATA%\Claude\claude_desktop_config.json. Restart the app after editing.

2. Install the agent skill (Cursor, recommended) so the model knows when to use extract vs fetch and how to design schemas:

cp -r docs/searxng-skill ~/.cursor/skills/searxng-capabilities

Reload Cursor. Workflows: docs/searxng-skill/SKILL.md.

MCP implementation details: mcp-server/README.md. Extract pipeline: extractor-sidecar/README.md, docs/extract-sidecar-specification.md.

Operations

Services use restart: unless-stopped. Redis data lives in a Docker volume; SearXNG config in ./searxng/.

docker compose ps
docker compose logs --tail=50
docker compose pull && docker compose up -d   # update images

Logs rotate at 1MB per container. Tune engines in searxng/settings.yml if needed.

Troubleshooting

Full reset (removes volumes):

docker compose down -v && docker compose up -d

Documentation

Acknowledgments

Built on SearXNG, Docker, Redis, and the Model Context Protocol.