Skip to content

SearXNG Automation

Patterns and tools for using SearXNG as a programmatic search provider.

What it is

SearXNG Automation involves interacting with a self-hosted SearXNG instance via its JSON API to provide web search capabilities to AI agents, scripts, and other services.

What problem it solves

It provides a private, rate-limit-friendly way for local scripts and AI models to access web information. Unlike commercial search APIs (like Google or Bing), SearXNG is free to use once self-hosted and aggregates results from dozens of sources.

Where it fits in the stack

Category: Services / Search Automation. It acts as the "Web Retrieval" layer for AI agents and automated research pipelines.

Typical use cases

  • Giving a local LLM (via Ollama or LangChain) the ability to search the live web.
  • Automated brand monitoring or news aggregation scripts.
  • Building a private "Daily Briefing" that searches for specific topics every morning.
  • Programmatically checking for broken links or updated information on specific sites.

Strengths

  • No API Keys: Once hosted, you have full control and no per-query costs.
  • Aggregated Data: Access results from Google, Bing, Wikipedia, and 70+ others in one request.
  • Privacy: Your automated queries are proxied and stripped of tracking data.
  • JSON Output: Results are returned in a clean, machine-readable format.

Limitations

  • Upstream Blocking: High-frequency automation can lead to your SearXNG IP being blocked by major search engines (use proxies if needed).
  • Format Stability: Changes in upstream engine HTML can occasionally break scrapers, requiring SearXNG updates.

When to use it

  • When building AI agents that need to browse the web without expensive API fees.
  • When you want to maintain full privacy for your automated search queries.
  • For niche research tasks that require data from multiple search engines simultaneously.

When not to use it

  • For extremely high-volume (thousands per minute) search tasks without a sophisticated proxy setup.
  • If you need real-time, millisecond-latency search results (SearXNG latency is tied to the slowest upstream engine).

Getting started

Prerequisites

  • A running SearXNG instance.
  • JSON output enabled in settings.yml: 
    search:
  formats:
    - html
    - json

Hello World (curl)

Test the API from your command line: 

curl "http://localhost:8080/search?q=open+source+llm&format=json"

Hello World (Python)

import requests

def search_searxng(query):
    url = "http://localhost:8080/search"
    params = {"q": query, "format": "json"}
    response = requests.get(url, params=params)
    return response.json()

results = search_searxng("Model Context Protocol")
print(f"Top Result: {results['results'][0]['title']}")

CLI examples

Automation often involves filtering results via command line tools like jq.

# Get only the URLs of the top 5 results
curl -s "http://localhost:8080/search?q=homelab&format=json" | jq -r '.results[:5][].url'

# Search specifically for images and save the first URL to a file
curl -s "http://localhost:8080/search?q=sunset&categories=images&format=json" | jq -r '.results[0].img_src' > image_url.txt

# Search Wikipedia via SearXNG
curl -s "http://localhost:8080/search?q=Python&engines=wikipedia&format=json" | jq -r '.results[0].content'

API examples

LangChain Integration

SearXNG is a first-class tool in the LangChain ecosystem.

from langchain_community.utilities import SearxSearchWrapper

# Configure the wrapper
search = SearxSearchWrapper(searx_host="http://localhost:8080")

# Run a query
output = search.run("What are the latest features of n8n?")
print(output)

n8n Integration (HTTP Request)

In n8n, use the HTTP Request node to fetch search results: - Method: GET - URL: http://searxng:8080/search - Query Parameters: - q: {{ $json["query"] }} - format: json - This allows your workflows to "research" topics before making decisions or sending notifications.

  • SearXNG (The core service)
  • n8n (To orchestrate search-based workflows)
  • Ollama (To process search results with local AI)
  • LangChain (For building search-enabled agents)
  • Tavily (Commercial alternative)

Sources / References

Contribution Metadata

  • Confidence: high
  • Last reviewed: 2026-06-12