ddg-search

name: ddg-search description: DuckDuckGo HTML search scraper CLI with JSON, CSV, OpenSearch, markdown, and compact outputs. homepage: https://github.com/camohiddendj/ddg-search metadata: { "openclaw": { "emoji": "🦆", "requires": { "bins": ["ddg-search"] }, "install": [ { "id": "node", "kind": "node", "package": "ddg-search", "bins": ["ddg-search"], "label": "Install ddg-search CLI (npm)", }, ], },

}

Search DuckDuckGo from the command line. Results go to stdout; progress goes to stderr.

Quick reference

ddg-search "query"                          # default: JSON, 5 pages
ddg-search -f compact "query"               # minimal-token output (best for LLM context)
ddg-search -f jsonl "query"                 # one JSON object per line
ddg-search -n 10 "query"                    # stop after 10 results
ddg-search -p 2 -f json "query"             # 2 pages, JSON
ddg-search -r us-en -t w "recent topic"     # US-English, past week
ddg-search -p 0 "query"                     # unlimited pages (scrape all)

Options

Flag	Long	Description	Default
`-f`	`--format`	Output format: `json`, `jsonl`, `csv`, `opensearch`, `markdown`, `compact`	`json`
`-p`	`--pages`	Max pages to scrape (0 = unlimited)	`5`
`-n`	`--max-results`	Stop after this many results	all
`-r`	`--region`	Region code (e.g. `us-en`, `uk-en`)	all regions
`-t`	`--time`	Time filter: `d` (day), `w` (week), `m` (month), `y` (year)	none

Choosing a format

compact: Use for feeding results into an LLM. Minimal tokens, no JSON overhead.
jsonl: Use when piping to line-oriented tools or streaming processors.
json: Use when you need structured data with OpenSearch metadata, zero-click answers, and spelling corrections. Pipe through jq for field extraction (e.g. | jq '.items[].link').
csv: Use for spreadsheets or tabular analysis.
markdown: Use for human-readable output or embedding in documents.
opensearch: Use when producing Atom XML feeds.

Extracting URLs from JSON output

ddg-search "query" | jq -r '.items[].link'

Notes

DuckDuckGo may trigger bot detection. The tool stops early and returns whatever results were collected.
Random delays (800–2900 ms) are inserted between page fetches automatically.
Progress messages appear on stderr, so redirecting stdout captures only results.