SearXNG Research Stack

Zero-cost 8-engine web search for AI agents. Designed as a Serper (Google) complement for market research — surfaces competitors, distributors, and industry content that Google SEO buries.

What this gives you

vs plain Google/Serper	Why it matters
4 independent web indexes	Brave, Mojeek, Presearch, Marginalia each rank differently
No SEO bubble	Finds non-optimized industry pages Google hides
Cross-validation	Same query across multiple indexes = higher confidence
Zero per-query cost	Self-hosted, no API billing

For Humans

One-command setup

git clone https://github.com/brandnewmax/searxng-research-stack.git
cd searxng-research-stack
bash setup.sh

That's it. The script checks Docker/Python, installs deps, starts SearXNG, waits for it to be healthy, and registers the MCP server with Claude Code.

Prerequisites: Docker Desktop, Python 3.9+.

Manual setup (if you prefer)

git clone https://github.com/brandnewmax/searxng-research-stack.git
cd searxng-research-stack
pip install httpx mcp
docker compose up -d
claude mcp add searxng-search -e SEARXNG_URL=http://localhost:8080 -- python3 $(pwd)/mcp-server/server.py

For Claude Code (self-install)

If you are Claude Code and a user has pointed you at this repo, run:

bash setup.sh

Then verify:

claude mcp get searxng-search

If Docker is not installed, tell the user to install Docker Desktop first. If claude mcp add fails, show the user the manual setup command from the section above.

MCP Tools

search — Multi-engine web search

query: Germany beauty equipment distributor medical aesthetic clinic

Returns structured results with engine labels ([brave], [mojeek], [presearch]), relevance scores, and which specific engines surfaced each result.

When to use: competitor discovery, distributor hunting, finding pages that don't rank on Google.

search_compare — Cross-validation mode

Same query, results grouped by engine. Diff the [brave] column against Serper to spot what Google missed.

When to use: verifying R1 conclusions against independent sources.

search_and_rerank — Semantic reranking

SearXNG raw results → Jina Reranker re-ranks by semantic relevance. Requires JINA_API_KEY.

When to use: deep research, ambiguous queries where keyword match isn't enough.

check_engines — Health check

Shows which engines are active and which are rate-limited.

Usage pattern (with Serper)

Market size / trends  → Serper (Google wins here)
Competitor discovery  → search_compare (both, compare)
Distributor hunting   → search (SearXNG wins here)
Cross-validation      → search_compare (diff the results)
Deep dive             → search_and_rerank (semantic ranking)

Engines

Engine	Type	Unique value
Brave	30B+ independent index	Different ranking from Google
Mojeek	UK independent index	Strong on B2B/industrial content
Presearch	Decentralized	Community-curated ranking
Marginalia	Non-commercial sites	Finds niche players without SEO
Wikipedia	Knowledge base	Industry definitions, company backgrounds
Wikidata	Structured data	Machine-readable facts
Google News	Real-time news	Industry updates, trade shows
Bing News	Real-time news	Alternative news ranking

Troubleshooting

Docker not running?

open -a Docker  # macOS

No results from engines?

curl "http://localhost:8080/search?q=test&format=json" | python3 -c "import sys,json; d=json.load(sys.stdin); print('Failed engines:', d.get('unresponsive_engines',[]))"

Rate limits reset automatically after 2-5 minutes. The MCP server has a 5-minute cache and 3-second throttle to minimize this.

Container won't start?

docker compose down && docker compose up -d

Cost

$0. Serper is fixed-cost via MCP. SearXNG runs on your machine. Jina Reranker (optional) costs ~$0.045 per million tokens — less than $0.02 for a complete research project.