--- title: "SearXNG review: how we actually run it" description: "A free, self-hosted metasearch engine. It asks other search services your query and returns one merged list, as a web page or as JSON." canonical: "https://leadlinepartners.co/tools/searxng" tier: "In our stack" category: "Scraping & Custom Data" last_reviewed: "2026-10-31" publisher: "Lead Line Partners" --- # SearXNG A free, self-hosted metasearch engine. It asks other search services your query and returns one merged list, as a web page or as JSON. In our stack. Last reviewed October 31, 2026. ## At a glance - Website: https://docs.searxng.org/ - Best for: AI-native, Agencies - Ease of use: Hard (A Docker container, a settings file and a request throttle to keep upstream engines from banning you: trivial for a builder, a wall for most sales teams.) - Interface: API, agent-ready - Agent readiness: With JSON output on, it answers a plain HTTP request on the local machine with JSON, so an agent can call it from a short script with no key. - Pricing: Free and open source. The cost is a machine that stays on, upkeep and ban risk. ## Who it's for A keyless local JSON endpoint suits agent-run research, and with no per-query fee an agency can run it across every client's research without metering. ## About What SearXNG actually is. SearXNG is a free metasearch engine. You send it one query, it asks a spread of other search services, merges what comes back and hands you a single list. It doesn't track or profile the person searching. The project began in mid-2021 as a fork of the older searx and is licensed AGPL-3.0. You run it yourself, usually as a Docker container, so the search happens on your own machine instead of inside someone else's account. That's the appeal and the catch. No key. No per-query fee. But it works by reading other engines' result pages, and those engines push back when queries arrive too fast, so plan for a throttle, a machine that stays on, and updates you have to read before you apply them. ### What it does - Pulls results from up to 261 search services, which you switch on or off in a settings file. - Returns results as a web page or as JSON, once JSON output is enabled in the settings. - Doesn't track or profile the people using it. - Ships as a Docker image, with Docker Compose as the recommended route and a single container as the manual option. - Lists about 70 maintained public instances on searx.space if you'd rather not host one yourself. - Works over Tor and offers its interface in 58 languages. - Updates by pulling a newer image and restarting, and the docs tell you to re-read the changed templates and settings each time. We haven't measured its recall against a paid search API, so treat it as a first pass and open the page before you rely on anything it returns. ## Runs well with The tools we'd run next to SearXNG, and what each pairing covers. - [Claude Code](https://leadlinepartners.co/tools/claude-code): An agent calls the search script, reads the JSON and decides which links are worth opening. - [Firecrawl](https://leadlinepartners.co/tools/firecrawl): SearXNG returns links and snippets only, so a page reader like Firecrawl is what turns a chosen link into text you can cite. ## How we run it Its job inside client systems. SearXNG is first in our web search order and has been since 7 Oct 2026. It runs as one Docker container on a single machine, reachable only from that machine, with JSON output on and the Google engine off. A small Python wrapper sends each query to it, holds the whole machine to one query every 3 seconds, retries once after 45 seconds if no engine answers, and prints one JSON object per query: title, link, snippet and which engines replied. Our research work reads that JSON. A result is a pointer, never a source, so we open the page before we cite anything. ## Pricing From the SearXNG pricing page, verified October 31, 2026. | Plan | Price | Includes | | --- | --- | --- | | Self-hosted (highlighted) | $0 (free, no account, no per-query fee) | Free and open source under AGPL-3.0; No vendor account, no API key, no per-query fee; You supply a machine that stays on and runs Docker; You do the upkeep: pull new images and re-read changed settings on update; Upstream engines can serve CAPTCHAs or suspend your IP if queries go out too fast | The dollar cost is zero. The real cost is a machine that stays on (ours is a laptop, so search only works while it's awake), the time it takes to pull new images and re-read changed settings, and attention to how fast queries go out. Bursts are what trigger bans from the upstream engines, not total volume. Our throttle works out at about 1,200 queries an hour. We haven't logged an update cadence, so we can't tell you how often upkeep comes round. ## The verdict ### Choose it if - You run research through scripts or agents and want search with no key, no account and no per-query fee. - You're comfortable keeping a Docker container up and re-reading its settings when you update it. - You can throttle your own queries, because bursts are what make upstream engines push back. ### Skip it if - You want search with no machine to maintain. A hosted search API takes that job off you. - You need guaranteed recall or an uptime promise. We haven't measured recall, and single engines can go quiet mid-run. - Nobody on your team is comfortable with Docker or a settings file. ## Alternatives If not SearXNG, then what. - [Exa.ai](https://leadlinepartners.co/tools/exa-ai) (Free tier (20,000 requests/mo) then pay-as-you-go, e.g. Search $7/1k requests; enterprise quote available): Choose Exa when you'd rather pay for a hosted search API than keep a container up and manage engine bans. - [Apify](https://leadlinepartners.co/tools/apify) (Free tier ($5/mo usage credit) then paid plans from $29/mo (Starter), usage-based overage billing): Choose Apify when you need a ready-made scraper for one specific site, not a general web search. ## FAQ Asked in real client work. ### Is SearXNG really free? Yes. It's open source under AGPL-3.0, with no account, no key and no per-query fee. What you pay in is a machine that stays on, upkeep when you update, and care about how fast you send queries. ### Does SearXNG have an API? Once JSON output is turned on in the settings file, its search endpoint returns JSON when you ask for it. We call it from a short Python script. We haven't checked whether the project ships an MCP server or an official command line tool, so we don't claim one. ### Why did you add a throttle? On 7 Oct 2026 four research lanes ran at full speed. DuckDuckGo and Qwant started serving CAPTCHAs, and Brave and Google suspended us for 180 seconds. Bursts were the problem, not volume. We now send one query every 3 seconds across the machine. ### Can I trust what it returns? Treat each result as a pointer, not a source. The page you land on sets how much weight a claim can carry, and a snippet is never evidence. ### Is it as good as a paid search API? We haven't measured, so we can't give you a comparison. We keep a paid option on pause and would revisit it only if recall proved too thin. ### How do I run it? The docs recommend Docker Compose: download the compose file and the example environment file from the repo's container folder, copy the example to a real one, run docker compose up -d, then edit the core settings file. A single docker run also works, mapping a host port to the container's port 8080 and mounting config and data folders. If you expose it beyond your own machine, the docs advise a reverse proxy. --- We build and run SearXNG inside client GTM systems. Talk to us: https://leadlinepartners.co/contact How we classify tools: https://leadlinepartners.co/tools/methodology Read this on the web: https://leadlinepartners.co/tools/searxng Lead Line Partners, https://leadlinepartners.co