Bright Data CLI
Scrape, search, and extract structured web data — directly from your terminal.
Overview
@brightdata/cli is the official npm package for the Bright Data CLI. It installs the brightdata command (with bdata as a shorthand alias) for access to the full Bright Data API surface:
| Command |
What it does |
brightdata scrape |
Scrape any URL — bypasses CAPTCHAs, JS rendering, anti-bot protections |
brightdata search |
Google / Bing / Yandex search with structured JSON output |
brightdata discover |
AI-powered web discovery - find and rank results by intent with optional full-page content |
brightdata scraper create |
Build a Bright Data scraper from a natural-language description using AI |
brightdata scraper run |
Run a Bright Data scraper on a URL and return the data |
brightdata scraper heal |
Fix an existing scraper in place via AI self-healing (stops at an approval gate) |
brightdata scraper approve |
Approve (or reject) a self-healing fix that is awaiting approval |
brightdata pipelines |
Extract structured data from 40+ platforms (Amazon, LinkedIn, TikTok…) |
brightdata browser |
Control a real browser via Bright Data's Scraping Browser — navigate, snapshot, click, type, and more |
brightdata zones |
List and inspect your Bright Data proxy zones |
brightdata budget |
View account balance and per-zone cost & bandwidth |
brightdata skill |
Install Bright Data AI agent skills into your coding agent |
brightdata add mcp |
Add the Bright Data MCP server to Claude Code, Cursor, or Codex |
brightdata config |
Manage CLI configuration |
brightdata init |
Interactive setup wizard |
Table of Contents
Installation
Requires Node.js ≥ 20
macOS / Linux
curl -fsSL https://cli.brightdata.com/install.sh | sh
Windows
npm install -g @brightdata/cli
npm install -g @brightdata/cli
You can also run without installing:
npx --yes --package @brightdata/cli brightdata <command>
Quick Start
# 1. Run the interactive setup wizard
brightdata init
# 2. Scrape a page as markdown
brightdata scrape https://example.com
# 3. Search Google
brightdata search "web scraping best practices"
# 4. Extract a LinkedIn profile
brightdata pipelines linkedin_person_profile "https://linkedin.com/in/username"
# 5. Check your account balance
brightdata budget
# 6. Install the Bright Data MCP server into your coding agent
brightdata add mcp
Authentication
Get your API key from brightdata.com/cp/setting/users.
# Interactive — opens browser, saves key automatically
brightdata login
# Via GitHub CLI — no browser needed (requires gh)
brightdata login --github
# Non-interactive — pass key directly
brightdata login --api-key <your-api-key>
# Environment variable — no login required
export BRIGHTDATA_API_KEY=your-api-key
On first login the CLI checks for required zones (cli_unlocker, cli_browser) and creates them automatically if missing.
# Clear saved credentials
brightdata logout
brightdata add mcp uses the API key stored by brightdata login. It does not currently read BRIGHTDATA_API_KEY or the global --api-key flag, so log in first before using it.
Free Tier
Every new Bright Data account includes a recurring monthly free tier — no credit card or commitment required to start. See the Free Tier docs for full details.
5,000 free credits per month (≈ $7.50), renewing on the 1st of each month. Unused credits don't roll over.
These credits cover the products most CLI commands use:
| Product |
CLI commands |
Credit cost |
| Unlocker API |
scrape |
1 credit / request |
| SERP API |
search |
1 credit / request |
| Web Scraper API |
pipelines |
1 credit / request |
| Scraper Studio |
scraper create / run / heal |
1 credit / page load (shared pool, not per record) |
Not included in the monthly free credits: proxy products (Datacenter, ISP, Residential) and the Browser API (brightdata browser). These get a separate one-time $2 trial (valid 7 days), plus a $5 bonus when you add a payment method (valid 30 days).
Adding a credit card is a verification step only — you are not charged unless your free credits are exhausted and you have funds deposited. Accounts on custom pay-as-you-go or pre-commit subscription plans don't receive the recurring monthly free credits.
Check your remaining balance any time with brightdata budget.
Commands
init
Interactive setup wizard. The recommended way to get started.
brightdata init
Walks through: API key detection → zone selection → default output format → quick-start examples.
| Flag |
Description |
--skip-auth |
Skip the authentication step |
-k, --api-key <key> |
Provide API key directly |
scrape
Scrape any URL using Bright Data's Web Unlocker. Handles CAPTCHAs, JavaScript rendering, and anti-bot protections automatically.
brightdata scrape <url> [options]
| Flag |
Description |
-f, --format <fmt> |
markdown · html · screenshot · json (default: markdown) |
--country <code> |
Geo-target by ISO country code (e.g. us, de, jp) |
--zone <name> |
Web Unlocker zone name |
--mobile |
Use a mobile user agent |
--async |
Submit async job, return a snapshot ID |
-o, --output <path> |
Write output to file |
--json / --pretty |
JSON output (raw / indented) |
-k, --api-key <key> |
Override API key |
Examples
# Scrape as markdown (default)
brightdata scrape https://news.ycombinator.com
# Scrape as raw HTML
brightdata scrape https://example.com -f html
# US geo-targeting, save to file
brightdata scrape https://amazon.com -f json --country us -o product.json
# Pipe to a markdown viewer
brightdata scrape https://docs.github.com | glow -
# Async — returns a snapshot ID you can poll with `status`
brightdata scrape https://example.com --async
search
Search Google, Bing, or Yandex via Bright Data's SERP API. Google results include structured data (organic results, ads, people-also-ask, related searches).
brightdata search <query> [options]
| Flag |
Description |
--engine <name> |
google · bing · yandex (default: google) |
--country <code> |
Localized results (e.g. us, de) |
--language <code> |
Language code (e.g. en, fr) |
--page <n> |
Page number, 0-indexed (default: 0) |
--type <type> |
web · news · images · shopping (default: web) |
--device <type> |
desktop · mobile |
--zone <name> |
SERP zone name |
-o, --output <path> |
Write output to file |
--json / --pretty |
JSON output (raw / indented) |
-k, --api-key <key> |
Override API key |
Examples
# Formatted table output (default)
brightdata search "typescript best practices"
# German localized results
brightdata search "restaurants berlin" --country de --language de
# News search
brightdata search "AI regulation" --type news
# Page 2 of results
brightdata search "web scraping" --page 1
# Extract just the URLs
brightdata search "open source scraping" --json | jq -r '.organic[].link'
# Search Bing
brightdata search "bright data pricing" --engine bing
discover
AI-powered web discovery. Submit a query with optional intent, and Bright Data finds, ranks, and optionally extracts full-page content for each result.
brightdata discover <query> [options]
| Flag |
Description |
--intent <text> |
AI intent to evaluate and rank result relevance |
--country <code> |
ISO country code (default: US) |
--city <name> |
City for localized results (e.g. "New York") |
--language <code> |
Language code (default: en) |
--num-results <n> |
Number of results to return |
--filter-keywords <words> |
Comma-separated keywords that must appear in results |
--include-content |
Include full page content in each result |
--no-remove-duplicates |
Keep duplicate results |
--start-date <date> |
Only content updated from date (YYYY-MM-DD) |
--end-date <date> |
Only content updated until date (YYYY-MM-DD) |
--timeout <seconds> |
Polling timeout (default: 600) |
-o, --output <path> |
Write output to file |
--json / --pretty |
JSON output (raw / indented) |
-k, --api-key <key> |
Override API key |
Examples
# Basic discovery — table output
brightdata discover "AI trends"
# With AI intent for relevance ranking
brightdata discover "AI trends" \
--intent "Prioritize institutional reports for VC research"
# Include full page content as markdown
brightdata discover "AI trends" --include-content --num-results 5
# Geo-targeted with date range
brightdata discover "best restaurants" --country US --city "New York" \
--start-date 2025-01-01 --end-date 2025-12-31
# Filter results by keywords
brightdata discover "generative AI SaaS" --filter-keywords "revenue,SaaS"
# JSON output to file
brightdata discover "AI trends" --num-results 10 --pretty -o results.json
# Pipe-friendly — redirected stdout outputs JSON automatically
brightdata discover "AI trends" --include-content --num-results 3 > results.json
scraper create
Build a Bright Data scraper from a natural-language description using AI.
brightdata scraper create <url> <description> [options]
| Flag |
Description |
--name <name> |
Scraper template name (default: cli-scraper-<timestamp>) |
--deliver-webhook <url> |
Webhook URL for the deliver stub (default: https://example.com/webhook) |
--timeout <seconds> |
Polling timeout in seconds (default: 600) |
--max-retries <n> |
Max retries on the AI-Flow concurrent-job cap 429 (default: 4). See below. |
--no-retry |
Fail immediately on 429 instead of waiting. Same as --max-retries 0. |
-o, --output <path> |
Write the JSON envelope to a file (see below) |
--json / --pretty |
JSON output (raw / indented) |
--legacy-output |
Write the pre-v0.3 bare AI-progress payload to -o instead of the envelope. Migration only. |
--timing |
Show request timing |
-k, --api-key <key> |
Override API key |
Note: The scraper is created with a placeholder webhook delivery target (https://example.com/webhook). You can reconfigure the actual delivery endpoint in the Bright Data web UI after creation.
Output envelope (-o create.json)
Every termination path — success or failure — writes the same JSON envelope shape:
{
"collector_id": "c_mp7x8a9b2c0d1e2f",
"name": "my-product-scraper",
"status": "done",
"completed_steps": ["prepare_intent_analyzer", "planner", "..."],
"view_url": "https://brightdata.com/cp/scrapers/c_mp7x8a9b2c0d1e2f",
"created_at": "2026-05-18T07:28:30Z"
}
On failure paths the envelope adds an error field and the status reflects the failure category (ai_trigger_failed, failed, poll_failed). The collector_id and view_url are still present so you can recover or inspect the half-built scraper.
This makes the documented chain in recipes.md work as written:
brightdata scraper create https://example.com/product/1 "..." \
-o create.json
COLLECTOR_ID=$(jq -r '.collector_id' create.json)
brightdata scraper run "$COLLECTOR_ID" https://example.com/product/2
The file format follows the -o extension, so .json is written compact (ideal for jq). Use --pretty for indented JSON on stdout when you omit -o.
Use --legacy-output if you have an existing script that depended on the pre-v0.3 bare-progress shape; the flag is supported for one minor version while you migrate.
Concurrent-job cap & auto-backoff
The Bright Data AI Flow caps concurrent scraper create generations per account (currently 3). If you exceed it, the API returns 429 Cannot run more than N jobs in parallel. The CLI handles this automatically: it waits with exponential backoff + jitter and retries up to --max-retries times (default 4). During the wait the CLI prints status lines so you know it isn't hung:
Triggering AI generation...
Hit AI-Flow concurrent-job cap (429). Waiting 32s before retry 1/4...
Hit AI-Flow concurrent-job cap (429). Waiting 67s before retry 2/4...
Generating scraper...
If the cap is still hit after all retries, the CLI exits with a stderr note pointing at the half-built collector's dashboard URL so you can inspect or delete it manually (Bright Data does not yet expose programmatic collector deletion).
Use --no-retry if you want the old fail-fast behavior — typically for scripts that prefer to handle backoff themselves.
Examples
# Build a scraper for a product page
brightdata scraper create https://example.com/product/1 \
"Extract title, price, and image URL from this product page"
# Name the scraper and save the envelope to a file
brightdata scraper create https://example.com/product/1 \
"Extract title, price, and image URL from this product page" \
--name my-product-scraper -o create.json
# Capture the collector_id for chaining
COLLECTOR_ID=$(jq -r '.collector_id' create.json)
# Fan out 10 parallel creates — the CLI serialises automatically via 429 backoff
for url in $(cat urls.txt); do
brightdata scraper create "$url" "Extract title, price, ..." \
--name "scraper-$(basename $url)" &
done; wait
# Disable the auto-backoff (fail fast on 429)
brightdata scraper create https://example.com/product/1 \
"Extract title, price, and image URL from this product page" \
--no-retry
# Use a custom webhook delivery URL
brightdata scraper create https://example.com/product/1 \
"Extract title, price, and image URL from this product page" \
--deliver-webhook https://my-app.com/ingest
scraper run
Run a scraper (built with scraper create or in the web UI) against one or more URLs and get the extracted data.
brightdata scraper run <collector_id> [url] [options]
Provide URLs in exactly one of three ways:
- Positional
<url> — single URL (legacy form, unchanged).
--urls <u1,u2,...> — comma-separated list.--input-file <path> — file with one URL per line, or a JSON array of URL strings, or a JSON array of {"url": "..."} objects.
| Flag |
Description |
--urls <list> |
Comma-separated list of URLs (multi-URL batch path) |
--input-file <path> |
File with URLs (txt one-per-line, or JSON array) |
--sync |
Use the synchronous /dca/crawl endpoint (single-URL only, server-side cap of 25–50s) |
--sync-timeout <seconds> |
Sync-mode server timeout, 25–50 (default: 50) |
--timeout <seconds> |
Polling timeout (default: 600 single-URL, 3600 batch) |
--name <name> |
Human-readable job name |
--version <version> |
Scraper version (e.g. dev) |
-o, --output <path> |
Write output to file |
--json / --pretty |
JSON output (raw / indented) |
--timing |
Show request timing |
-k, --api-key <key> |
Override API key |
Routing
- Single URL (positional, or one entry via
--urls / --input-file) → async flow: /dca/trigger_immediate → poll /dca/get_result. Use --sync for /dca/crawl (one-shot, 25–50s).
- Multiple URLs (
--urls / --input-file with 2+ entries) → single POST to /dca/trigger with an array body, one collection_id, polled via /dca/dataset. This mirrors the canonical batch shape used by the reference SDKs (triggerWithUrls / trigger_with_urls). --sync is incompatible with multi-URL — /dca/crawl accepts only a single URL.
If a single URL expands to more pages than the realtime job limit allows (paginated listings, infinite scroll), the CLI automatically falls back to the batch endpoint and prints a one-line notice. No flag required.
Examples
# Default: async + poll until results arrive
brightdata scraper run c_mp3tuab31lswoxvpws https://www.amazon.com/dp/B08N5WRWNW
# Save pretty-printed results to a file
brightdata scraper run c_mp3tuab31lswoxvpws https://www.amazon.com/dp/B08N5WRWNW \
--pretty -o product.json
# Sync mode for fast pages
brightdata scraper run c_mp3tuab31lswoxvpws https://example.com/p/1 --sync
# Sync with a shorter server timeout and a job name
brightdata scraper run c_mp3tuab31lswoxvpws https://example.com/p/1 \
--sync --sync-timeout 30 --name first-test
# Multi-URL batch — one API call, one snapshot, one merged result array
brightdata scraper run c_mp3tuab31lswoxvpws \
--urls "https://example.com/p/1,https://example.com/p/2,https://example.com/p/3" \
--pretty -o products.json
# Multi-URL from a file (one URL per line; # comments and blank lines skipped)
brightdata scraper run c_mp3tuab31lswoxvpws --input-file urls.txt -o products.json
# Multi-URL from a JSON array
echo '["https://example.com/p/1","https://example.com/p/2"]' > urls.json
brightdata scraper run c_mp3tuab31lswoxvpws --input-file urls.json
scraper heal
Fix an existing scraper in place when it ran but returned wrong, empty, or partial data. The collector_id stays the same — the scraper is improved, not replaced. This is the maintenance twin of scraper create: it triggers Bright Data's AI self-healing flow (POST /dca/collectors/{id}/refactor_template), then polls progress.
brightdata scraper heal <collector_id> "<prompt>" [options]
You are the detector. The CLI never decides on its own that a scraper is broken — you inspect the run output and decide. The <prompt> is required (max 1000 chars); name exactly what is wrong and what the correct output should be. Vague prompts produce vague heals.
| Flag |
Description |
--url <url> |
Verify target woven into the success next_step hint (not sent to the h |
Это официальный CLI от Bright Data, инструмент командной строки, который собирает, ищет и извлекает структурированные данные из веба прямо из терминала с помощью API Bright Data.