scrape Skill (Bright Data) for Claude: What It Does and How to Install It

Makes Claude choose a Bright Data scraper to pull structured data from a website, or build one in Scraper Studio.

Installs: 10.9K; on skills.sh
Stars: 271; on GitHub
Files: 7; besides SKILL.md
Updated: Oct 2026; last repo commit

This skill picks the Bright Data product for a scraping job, then runs it. Gate 1 uses a ready Web Scraper API scraper, checked first in a bundled table of 25 and then in the live catalogue. Gate 2 builds a new scraper in Scraper Studio when none fits and the job repeats, spans many pages with one layout or needs browser actions. A one-off job with no shared layout falls back to Web Unlocker through the fetch skill.

It needs a Bright Data account. The README says to start with the agent-onboarding skill, which installs the bdata command line tool and signs you in once. Web Scraper API jobs run asynchronously, and the trigger script's header says each delivered record is billed. Check the record count of the first run.

Good for

Analysts pulling profiles, products, reviews or prices from sites
Developers who need a scrape to repeat on a schedule
Teams without a Playwright or Selenium setup
Researchers collecting job or property listings

Install scrape

Claude Code

Copy this prompt.

Install the scrape skill in Claude Code. Run this command:

npx -y skills add brightdata/skills -y

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js, git.

Pick one route. The whole-set route installs all 11 Bright Data skills listed in the README. The README says the other ten assume the agent-onboarding skill ran once, because it installs the bdata command line tool (Node.js 20 or later) and signs you in. Without Node.js, the scripts and REST calls use BRIGHTDATA_API_KEY instead. The skill needs a Bright Data account.

From the project README. Source.

What's in the skill

When it loads

Use when the user wants to scrape or extract structured data from any site: profiles, posts, products, reviews, prices, followers, job or property listings. Also for building a scraper, crawling a site for data, discovery by keyword or category, asks that say "data fields" or "recurring", or that name a dataset_id, snapshot_id, Web Scraper API, or Scraper Studio. Not the page itself (use fetch), not a search query (use search), not cost or credits (use billing).

Sections in SKILL.md

The gates - run in order, first yes wins
The boundaries
The two override words
Web Scraper API is asynchronous by default
Example - one TikTok profile
Read next
Red flags - stop if you catch yourself doing one of these

Key facts

Author: brightdata (repo owner)
License: MIT (crawl data; the README badge also says MIT)
Installs on skills.sh: 10,881 (Nov 2, 2026)
GitHub stars: 271 for the whole brightdata/skills repo (Nov 2, 2026)
Last repo commit: Oct 29, 2026
What it includes: 8 files including SKILL.md: 4 reference files and 3 scripts (find-scraper.mjs, trigger.mjs, poll.mjs)
Needs: A Bright Data account. The scripts read BRIGHTDATA_API_KEY or the bdata CLI login file, and run on Node.js 18 or newer
Part of: The Bright Data Skills repo, which the README lists as 11 skills (agent-onboarding, scrape, instagram, datasets, fetch, search, browser, billing and three interface guides)
Works in: Claude Code, Codex, Cursor and other agents that read SKILL.md files, per the README

What it includes

Already-collected data goes to the datasets skill
If the data is a huge generic corpus, needed once, and a few months old is fine, the datasets skill handles it instead of scraping.
Read · Gate 0: marketplace check
Ready scrapers from a bundled top-25 table or the live catalogue
Checks the table (YouTube, Facebook, TikTok, Instagram, LinkedIn, Amazon, Reddit and more) before spending an API call. Inputs are not always URLs.
Try: “Get the follower count and bio for tiktok.com/@nasa.”
Network · Gate 1: Web Scraper API
Look up a scraper by gd_ id, domain or name
Reads the scrapers catalogue and prints inputs and outputs. It caches the catalogue in your temp folder, and SKILL.md says one empty-body probe can start a billed job.
Runs code · find-scraper.mjs
Start one collection job and print the snapshot id
Sends one input to a scraper. Its header says it spends money, billed per delivered record, and --limit sets limit_per_input.
Runs code · trigger.mjs
Turn a snapshot or job id into records
Polls with backoff and prints records, or writes them to a file with --out.
Runs code · poll.mjs
Build, run, heal and schedule a scraper
Uses bdata scraper create, run, heal and approve. Scheduling is set in the Bright Data control panel.
Try: “Build a scraper for this news site that runs daily and returns title, url and author.”
Runs code · Gate 2: Scraper Studio
One-off fetch for pages with no shared layout
Hands back to the fetch skill, where you own the parser. Four error codes can refuse a target.
Network · Fall-through: Web Unlocker
"fields" forces structured output; "recurring" adds a schedule
Two separate words for output and schedule. Neither skips the gates.
Read · Override words
Sends Instagram, pages, searches and browser setups elsewhere
Instagram goes to the instagram skill, pages to fetch, queries to search and your own Playwright or Selenium code to browser.
Read · Boundaries

From the official docs

Permissions

Medium risk

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Triggering a scraper starts a billed job
trigger.mjs says a success queues a real collection and bills per delivered record. One URL can return many records, so check the first run and use --limit. github.com
The id probe can start a billed job
SKILL.md says find-scraper.mjs sends one empty-body probe for an id the catalogue omits, and that on a door with no required field it can start a billed job. It then prints the snapshot id and the cancel call. github.com
Auto-approve skips the heal review
scraper-studio.md says a heal stops at awaiting_approval by default, and that --auto-approve with --auto-save approves it and saves the template without review. Leave those flags off unless you trust the fix. github.com
Scripts read your Bright Data key or login file
They use BRIGHTDATA_API_KEY, or the CLI credentials.json under APPDATA, ~/Library/Application Support/brightdata-cli or ~/.config/brightdata-cli. Their headers say the key is never printed. github.com
The README names Node.js and a CLI install
The README says agent-onboarding installs the bdata CLI (Node.js 20 or later). That install is a separate step from this skill. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. find-scraper.mjs only reads the catalogue, but trigger.mjs starts billed jobs.
Talks to	api.brightdata.com (BRIGHTDATA_API_BASE must point to a brightdata.com host)
Sign-in and keys	The scripts read BRIGHTDATA_API_KEY, or the bdata CLI credentials.json. The CLI login is approved in a browser, per the README.
Scopes	Claude can run node scripts/find-scraper.mjs, trigger.mjs and poll.mjs from the skill folder Claude can run bdata commands for pipelines and Scraper Studio Scripts call api.brightdata.com with your API key or CLI login poll.mjs can write records to a file you name with --out

Questions

What does the scrape skill do?

It decides how to collect structured data from a site, without asking you which Bright Data product to use. It then runs a ready scraper, builds a Scraper Studio scraper, or falls back to a plain fetch.

How do I install the scrape skill?

Use the install box at the top of this page. The whole-set route is first because the README says the other Bright Data skills assume agent-onboarding ran once. It needs Node.js and git.

Do I need a Bright Data account?

Yes. The skill signs you in through the bdata command line tool or reads BRIGHTDATA_API_KEY. A refused call is sent back to the agent-onboarding skill.

How much does a scrape cost?

SKILL.md gives no prices. The trigger script says it spends money, billed per delivered record, and one URL can return many records. Check the record count of the first run.

How is it different from the fetch skill?

This skill returns structured data fields. SKILL.md sends the page itself (markdown, HTML or a screenshot) to the fetch skill, and any search query to the search skill. Both sit in the same repo.

Does it work in the Claude app?

The install box has no Claude app upload, because the skill runs Node.js scripts and the bdata command line tool on your machine. The README names Claude Code, Codex and Cursor.

Is it safe?

It runs three Node.js scripts and bdata commands that call Bright Data with your key and can start billed jobs. SKILL.md says its scripts never print the key, and Scraper Studio holds a fix for approval unless you pass --auto-approve. Read the first trigger command before you let it run (our advice).

Information

updated Nov 2, 2026

Repository: brightdata/skills
Folder: skills/scrape

brightdata/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 9

scrape SKILL.md · github.com
brightdata/skills README · github.com
scrape web-scraper-api.md · github.com
scrape scraper-studio.md · github.com
scrape web-unlocker.md · github.com
scrape snapshots-and-jobs.md · github.com
scrape scripts/trigger.mjs · github.com
scrape scripts/poll.mjs · github.com
scrape scripts/find-scraper.mjs · github.com