discover
Search the web and rank results by AI-driven relevance. Returns scored results with title, description, and URL. Supports intent-based ranking, geo-targeting, date filtering, and keyword filtering.
LABSCO SUMMARY
The structured feeds are the reason to pick this over a scraper.
Scraping an Amazon product page and parsing it is work that breaks every few weeks. The matching web_data_* tool returns clean JSON for the same URL, faster and more reliably — and when there is no feed for the site, the general scrape still handles the bot protection.
What it is
A web-access server on Bright Data's network, with 69 tools organised into groups so you load only what you need. Five are always on — search, scrape, their batch forms and a relevance-ranked discovery search — and the rest are structured data feeds for named platforms, browser automation, and AI extraction.
What you get
- Search Google, Bing or Yandex with a cursor for pagination, and up to 10 queries in one call (
search_engine,search_engine_batch) - Any URL as Markdown with bot protection and CAPTCHA handled, and up to 10 URLs in one call (
scrape_as_markdown,scrape_batch) - A relevance-ranked discovery search that scores results against a stated intent, with geo, date and keyword filtering (
discover) - Structured JSON feeds instead of scraping the same page: e-commerce for Amazon, Walmart, eBay, Best Buy, Etsy, Home Depot, Zara and Google Shopping — 11 tools in the
ecommercegroup - Social platforms — LinkedIn, Instagram, Facebook, TikTok, YouTube, X and Reddit — as 23 tools in the
socialgroup - Remote browser automation for pages that need clicking, typing, scrolling or a login, as 13 tools in the
browsergroup - Raw HTML, AI extraction to structured JSON with your own prompt, and session usage counts in the
advanced_scrapinggroup (scrape_as_html,extract,session_stats) - Smaller groups for the rest: Crunchbase, ZoomInfo, Google Maps reviews and Zillow (
business); npm and PyPI package data (code); Yahoo Finance, GitHub files, app stores, Booking.com; and how ChatGPT, Grok and Perplexity answer a given prompt (geo)
Requirements
- Nothing to install on the hosted path: add
https://mcp.brightdata.com/mcp?token=YOUR_API_TOKENto your client. New accounts get 5,000 free requests a month. - Locally it is
npx @brightdata/mcpwithAPI_TOKENset. - To load more than the five always-on tools, add groups with
GROUPSor single tools withTOOLS— as environment variables locally, or&groups=and&tools=on the hosted URL. - The
web_data_*feeds are billed per record returned and can produce large results, so use theirnum_of_commentsanddays_limitlimits; if one fails,scrape_as_markdownworks on the same URL. RATE_LIMIT,POLLING_TIMEOUT,BASE_TIMEOUTandBASE_MAX_RETRIESbound spend and latency.
Setup effort
One command plus a key — npx @brightdata/mcp, then supply credentials
USE CASES
01
Reading a page your own fetch tool cannot get
Scrape it as Markdown. Bot protection and CAPTCHAs are handled on their network, so the call that fails everywhere else returns content here.
02
Pulling product data without writing a parser
Use the feed for the platform rather than scraping the page — Amazon, Walmart, eBay, Etsy and the rest return structured JSON with price, availability and reviews already separated. Turn the feeds on with the ecommerce group: GROUPS=ecommerce locally, or &groups=ecommerce on the hosted URL. Billing is per record, so cap the review counts.
03
Collecting a list of pages in one go
Batch up to ten URLs or ten queries per call instead of looping with retries. It is the difference between a research pass and a script you have to babysit.
04
Finding sources on a topic, ranked by intent
Use the discovery search rather than a keyword query: it scores results against what you are actually trying to find, and filters by country, date and keyword. Duplicates come out ranked rather than repeated.
05
Getting past a page that needs interaction
Switch on the browser group — GROUPS=browser locally, or &groups=browser on the hosted URL — and drive the page: click, type, scroll, sign in, rather than trying to fetch a URL that only exists after three interactions.
06
Structured data from a site with no feed
Scrape it and convert with AI extraction, giving your own prompt for the fields you want. The extract tool is off by default: add the advanced_scraping group, or locally just this tool with TOOLS=extract. Slower than a feed and far quicker than writing a parser for a one-off.
07
Seeing how the assistants describe your brand
Collect how ChatGPT, Grok and Perplexity answer a given prompt, with the geo group switched on (GROUPS=geo locally, or &groups=geo on the hosted URL). It is the closest thing to a rank check for the surfaces that increasingly answer instead of linking.
08
Keeping the tool list to what you use
Five tools are always on: search, scrape, their batch forms and discover. Add bundles with GROUPS (or &groups= on the hosted URL) and single tools with TOOLS. A coding agent wanting npm and PyPI data needs one group; loading all 69 puts every tool definition on every turn.
5 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
scrape_as_markdown
Scrape a single webpage URL with advanced options for content extraction and get back the results in MarkDown language. This tool can unlock any webpage even if it uses bot detection or CAPTCHA.
scrape_batch
Scrape multiple webpages URLs with advanced options for content extraction and get back the results in MarkDown language. This tool can unlock any webpage even if it uses bot detection or CAPTCHA.
search_engine
Scrape search results from Google, Bing or Yandex. Returns SERP results in JSON or Markdown (URL, title, description),Ideal forgathering current information, news, and detailed search results.
search_engine_batch
Run multiple search queries simultaneously. Returns JSON for Google, Markdown for Bing/Yandex.
BEFORE YOU INSTALL
What to know first
Tools you get
5 tools when we connected; the project documents 69. Only search_engine, search_engine_batch, scrape_as_markdown, scrape_batch and discover are on by default. The rest arrive by enabling groups (GROUPS, or &groups= on the hosted URL), with TOOLS adding individual picks.
Requires
npx on PATH; API_TOKEN in your environment
INSTALLATION
Connect your client
Server setup
npx
TERMINAL
npx @brightdata/mcpHosted endpoint
ENDPOINT URL
https://mcp.brightdata.com/mcp?token=YOUR_API_TOKENConnect a client to this instead of installing anything. Replace YOUR_API_TOKEN in the address with your Bright Data API token, which you get from your Bright Data account settings. Nothing else is needed. Taken from this project's own documentation.
MCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"Bright Data": {
"command": "npx",
"args": [
"@brightdata/mcp"
],
"env": {
"API_TOKEN": "<your-api-token-here>"
}
}
}
}Claude Code
CLAUDE CODE · TERMINAL
claude mcp add --transport http brightdata "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"Cursor, Warp
JSON
{
"mcpServers": {
"brightdata": {
"url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
}
}
}VS Code
VS CODE · JSON
{
"servers": {
"brightdata": {
"type": "http",
"url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
}
}
}Windsurf
WINDSURF · JSON
{
"mcpServers": {
"brightdata": {
"serverUrl": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
}
}
}Gemini CLI
GEMINI CLI · JSON
{
"mcpServers": {
"brightdata": {
"httpUrl": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
}
}
}Zed
ZED · JSON
{
"context_servers": {
"brightdata": {
"url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
}
}
}
