Bright Data Skills

Teach any coding agent to get web data: scrape, fetch, search, browse, and watch the bill

Quick Start • Skills • Agent Onboarding • Scrape • Datasets • Fetch • Search • Browser • Billing • CLI, MCP, SDK

A skill is a folder with a SKILL.md and, where needed, reference files and scripts. Any agent that reads SKILL.md files can use these: Claude Code, Codex, Cursor, and others. Nothing here is tied to one agent.

With the skills installed, an agent can:

• Scrape any site for structured data with ready scrapers, or build a new scraper in Scraper Studio when none fits
• Find a ready dataset when the data is already collected, and hand the purchase to a person instead of scraping it
• Fetch a blocked page through Web Unlocker as markdown, HTML, or a screenshot
• Search Google and other engines and get structured JSON, including the AI answer engines
• Drive a cloud browser with the user's own Playwright, Puppeteer, or Selenium code
• Check balance and costs before and after every job
• Pick the right surface for the task: CLI for the terminal, MCP for agents that decide at run time, SDK for code the user keeps

Add the skills to your project, for any coding agent (needs Node.js and git):

npx -y skills add brightdata/skills -y

For one skill, add --skill <name>, for example --skill scrape. In Windows PowerShell, use npx.cmd if scripts are blocked. No Node? Clone this repo and copy the folders under skills/ into the folder where your agent reads skills (many agents use .agents/skills).

Then start with agent-onboarding: it installs the Bright Data CLI (needs Node.js 20 or later) and logs in once. The user approves in a browser, on this machine or any other device, and the key never touches the chat. Without Node it falls back to the REST API with BRIGHTDATA_API_KEY.

The entry point. It installs everything, handles the one-time login without ever pasting a key into chat, registers an account by email when the user has none, fixes the common Windows and PATH problems, and routes the task to the right skill. The other ten assume it ran once.

Structured data from any site. The skill checks what already exists before spending anything: a bundled top-25 table costs zero calls, and one free call searches the live scraper catalogue. When a ready scraper fits, it runs it. When none fits, Scraper Studio builds one from a plain-language prompt, heals it when the site changes, and runs single URLs synchronously or whole URL lists as one batch.

Not scraping. When the data is already collected and sold as a download, a huge generic corpus, needed once, where a few months old is fine, the agent checks coverage with free reads (plus, for an id the catalogue does not list, one probe that can start a billed job) and hands the purchase to a person in the control panel.

One URL in, unblocked page out through Web Unlocker: markdown for reading, HTML for parsing, or a screenshot, with bdata scrape.

Anything that starts from a query. SERP API for structured results (web, news, images and shopping tabs), the Google verticals with a scraper of their own (Maps, Shopping, Flights, Hotels, Images), the top-100 job for deep rank tracking in one request, and the AI answer engines (ChatGPT, Perplexity, Gemini, Google AI Mode) with their verified dataset ids.

For the user's own browser automation code. The skill hands Playwright, Puppeteer, or Selenium the right connect string for the cloud browser, which does the unblocking, and explains the errors that only show up in cloud sessions.

Balance, charges by product and by zone, and what a job will cost before it runs. REST first; the reference marks which facts were checked against the live cost APIs.

Three thin skills answer "which door do I use": brightdata-cli for one-off checks typed at a shell, brightdata-mcp for AI apps that pick tools at run time, and brightdata-sdk for code the user keeps and schedules.

MIT