Sandbox::https://docs.brightdata.com/ai/mcp-server/tools

Building an AI startup?

You might be eligible for our Startup Program. Get fully funded access to the infrastructure you’re reading about right now (up to $20K value).

Modes: Rapid (Free), Pro and four tool groups

Bright Data’s MCP server offers two modes to suit different needs:

Rapid (Free) - Scrape search results, unlock any public webpage as clean Markdown and get structured data from thousands of pre-built scrapers.
Pro - Access advanced scraping, AI extraction and full browser automation. Built for dynamic and large-scale use cases.
Groups - Pre-configured tool collections: scrapers (Scraper API tools), browser (browser automation), advanced_scraping (batch, HTML and AI extraction) and geo (ChatGPT and Perplexity answers). Older group IDs such as ecommerce, social and business still work and enable the scrapers tools.

To use Pro mode:

In Remote MCP, request Pro features by setting &pro=1.
In Local MCP, enable Pro features with PRO_MODE=true.

To use Groups:

In Remote MCP, request the Groups by setting &groups=<group_name>.
In Local MCP, enable Groups with GROUPS=<group_name>.

Rapid (Free) mode is enabled by default and is recommended for everyday browsing and data needs.

Bright Data MCP Tools Reference

Tool Name	Description	Group
`search_engine`	Scrape search results from Google, Bing, or Yandex. Returns SERP results in JSON for Google and Markdown for Bing/Yandex; supports pagination with the cursor parameter.	`all`
`scrape_as_markdown`	Scrape a single webpage with advanced extraction and return Markdown. Uses Bright Data’s unlocker to handle bot protection and CAPTCHA.	`all`
`search_engine_batch`	Run up to 10 search queries in parallel. Returns JSON for Google results and Markdown for Bing/Yandex.	`advanced_scraping`
`scrape_batch`	Scrape up to 10 webpages in one request and return an array of URL/content pairs in Markdown format.	`advanced_scraping`
`discover`	Search the web and rank results by AI relevance to an `intent`. Returns title, description, URL and relevance score; supports geo-targeting, date and keyword filters.	`all`
`scrape_as_html`	Scrape a single webpage with advanced extraction and return the HTML response body. Handles sites protected by bot detection or CAPTCHA.	`advanced_scraping`
`extract`	Scrape a webpage as Markdown and convert it to structured JSON using AI sampling, with an optional custom extraction prompt.	`advanced_scraping`
`session_stats`	Report how many times each tool has been called during the current MCP session.	`advanced_scraping`
`search_scrapers`	Find a pre-built scraper by site or data type, e.g. “amazon reviews” or “linkedin.com”. Returns `dataset_id`, name, domain and collection methods.	`scrapers`
`get_scraper_details`	Show a scraper’s description, input fields, `sample_input` and main output fields for one collection method. Call before `run_scraper`.	`scrapers`
`run_scraper`	Run a scraper method on a list of inputs shaped like `sample_input`. Waits up to 45 seconds; if the run is still going, returns a `snapshot_id` with status `running`. Billed per record.	`scrapers`
`get_scraper_progress`	Status of a scraper run: starting, running, ready, failed or canceled.	`scrapers`
`get_scraper_results`	Records of a finished scraper run. Returns status `running` if the run is not ready yet.	`scrapers`
`refresh_scrapers`	Reload the scraper list now instead of waiting for the daily refresh.	`scrapers`
`list_dataset_fields`	List the filterable fields of a searchable dataset with each field’s name, type and description. Call before `search_dataset` to learn which fields you can filter on.	`social`, `business`
`search_dataset`	Search a Bright Data dataset by a filter and get matching records back directly, with no snapshot step. Finds many records by criteria, unlike `run_scraper`, which collects fresh records by URL. Returns up to 10 records per call with `search_after` pagination.	`social`, `business`
`scraping_browser_navigate`	Open or reuse a scraping-browser session and navigate to the provided URL, resetting tracked network requests.	`browser`
`scraping_browser_go_back`	Navigate the active scraping-browser session back to the previous page and report the new URL and title.	`browser`
`scraping_browser_go_forward`	Navigate the active scraping-browser session forward to the next page and report the new URL and title.	`browser`
`scraping_browser_snapshot`	Capture an ARIA snapshot of the current page listing interactive elements and their refs for later ref-based actions.	`browser`
`scraping_browser_click_ref`	Click an element using its ref from the latest ARIA snapshot; requires a ref and human-readable element description.	`browser`
`scraping_browser_type_ref`	Fill an element identified by ref from the ARIA snapshot, optionally pressing Enter to submit after typing.	`browser`
`scraping_browser_fill_form`	Fill several form fields identified by ARIA refs in one call.	`browser`
`scraping_browser_screenshot`	Capture a screenshot of the current page; supports optional full_page mode for full-length images.	`browser`
`scraping_browser_network_requests`	List the network requests recorded since page load with HTTP method, URL, and response status for debugging.	`browser`
`scraping_browser_wait_for_ref`	Wait until an element identified by ARIA ref becomes visible, with an optional timeout in milliseconds.	`browser`
`scraping_browser_get_text`	Return the text content of the current page’s body element.	`browser`
`scraping_browser_get_html`	Return the HTML content of the current page; avoid the full_page option unless head or script tags are required.	`browser`
`scraping_browser_scroll`	Scroll to the bottom of the current page in the scraping-browser session.	`browser`
`scraping_browser_scroll_to_ref`	Scroll the page until the element referenced in the ARIA snapshot is in view.	`browser`
`chatgpt_ai_insights`	Send a prompt to ChatGPT and get the answer back as Markdown. Useful for GEO and LLM as a judge.	`geo`
`perplexity_ai_insights`	Send a prompt to Perplexity and get the answer back as Markdown. Useful for GEO and LLM as a judge.	`geo`

How do I get structured data from a site with the Bright Data MCP server?

Use the Scraper API tools to run any of thousands of pre-built scrapers and get structured JSON back. The tools are on by default in Rapid (Free) mode and in the scrapers group.

Call search_scrapers with a site or data type, for example “amazon reviews” or “linkedin.com”. It returns each matching scraper’s dataset_id and collection methods.
Call get_scraper_details with the dataset_id and one method. It returns the input fields and a sample_input.
Call run_scraper with an input list shaped like sample_input. A collect_by_url method returns one record per input. A discover_by_* method returns up to limit_per_input records per input (default 10). Every record is billed.
If run_scraper returns status running after 45 seconds, call get_scraper_progress until the run is ready, then call get_scraper_results.

Records with an error field explain inputs that failed, such as a URL of the wrong type. If a run fails, scrape_as_markdown works on the same URL as a fallback.

How do I search datasets by filter with the Bright Data MCP server?

Use list_dataset_fields and search_dataset to find many records that match your criteria, as opposed to run_scraper, which collects fresh records by URL. The search runs directly against the dataset with no snapshot step and returns up to 10 records per call. Under the hood, list_dataset_fields calls the Get dataset metadata endpoint and search_dataset calls the Search dataset endpoint of the Marketplace Dataset API. Three datasets are searchable:

`dataset_id`	Dataset
`gd_l1viktl72bvl7bjuj0`	LinkedIn people profiles
`gd_me5ppxjr2ge6icjuh0`	LinkedIn people profiles (contact-enriched)
`gd_l1vikfnt1wgvvqz95w`	LinkedIn company information

The workflow is two calls:

Call list_dataset_fields with a dataset_id. It returns each filterable field’s name, type and description.
Call search_dataset with the same dataset_id and a filter.

A filter is a tree with a maximum nesting depth of three. Each node is either a group, {"operator": "and" | "or", "filters": [...]}, or a leaf, {"name": ..., "operator": ..., "value": ...}. See filter syntax for the full list of leaf operators and their behavior. For example, this filter matches profiles in the United States whose position mentions data engineering:

{
  "operator": "and",
  "filters": [
    { "name": "country_code", "operator": "=", "value": "US" },
    { "name": "position", "operator": "includes", "value": "data engineer" }
  ]
}

The response contains hits (the matching records), total_hits and took (search time in milliseconds). To page through more than 10 results, set sort to "default" or a custom sort like [{"timestamp": "asc"}], then pass the response’s search_after cursor into the next call. Set sort to "random" to sample records instead.

Get started

Data extraction

Web search

AI training data

Proxy networks

Bright Data MCP tools

Building an AI startup?

Modes: Rapid (Free), Pro and four tool groups

Bright Data MCP Tools Reference

How do I get structured data from a site with the Bright Data MCP server?

How do I search datasets by filter with the Bright Data MCP server?

Building an AI startup?

​Modes: Rapid (Free), Pro and four tool groups

​Bright Data MCP Tools Reference

​How do I get structured data from a site with the Bright Data MCP server?

​How do I search datasets by filter with the Bright Data MCP server?

Modes: Rapid (Free), Pro and four tool groups

Bright Data MCP Tools Reference

How do I get structured data from a site with the Bright Data MCP server?

How do I search datasets by filter with the Bright Data MCP server?