Bright Data MCP Server: Setup for Claude, Cursor and Other Apps

• Searches Google or Bing and returns the results

  Runs a web search on Bing (default) or Google and returns the title, link and snippet for 10 results per page. You can set a country and a time range.

  Try: “Search for news from the last week about EU AI regulation.”

  Network · search_engine

• Reads any web page as clean text

  Opens a page through Bright Data's unblocking network, which handles bot checks and CAPTCHAs, and returns the content as markdown. It can also return plain text, HTML or JSON-LD.

  Try: “Read https://example.com/pricing and summarize the plans.”

  Network · scrape_as_markdown

• Runs several searches in one go

  Sends up to 5 search queries at once and returns the results for each.

  Try: “Search for these five competitor names and list the top result for each.”

  Network · search_engine_batch

• Reads several web pages in one go

  Fetches up to 5 URLs in a single call and returns each page as markdown, text, HTML or JSON-LD.

  Try: “Read these five product pages and compare their prices.”

  Network · scrape_batch

• Gets the raw HTML of a page

  Returns the unprocessed HTML source of a URL, fetched through the same unblocking network.

  Try: “Get the raw HTML of this page so I can check its meta tags.”

  Network · scrape_as_html

• Turns a web page into structured JSON

  Reads a page as markdown and then asks your AI client to convert it into JSON, optionally following your own extraction instructions.

  Try: “Extract the product name, price and rating from this page as JSON.”

  Network · extract

• Finds a ready-made scraper for a site

  Looks up Bright Data's catalog of over 1,200 prebuilt scrapers by site or data type, such as Amazon reviews or LinkedIn profiles.

  Try: “Is there a scraper for TikTok comments?”

  Network · search_scrapers

• Shows what a scraper needs and returns

  Describes one scraper method, including its input fields, an example input and the main fields it returns. It is meant to be called before running the scraper.

  Try: “What input does the Amazon product scraper take?”

  Network · get_scraper_details

• Runs a scraper and returns structured records

  Starts a prebuilt scraper with the inputs you give and waits up to 45 seconds for the records. If it is still running it returns an ID to check later. Each record returned is billed.

  Try: “Get the rating and review count for this Amazon product page.”

  Network · run_scraper

• Checks whether a scraper run has finished

  Reports whether a started scraper run is starting, running, ready, failed or canceled.

  Try: “Is my LinkedIn scrape finished yet?”

  Network · get_scraper_progress

• Fetches the records of a finished run

  Returns the records collected by a scraper run once it is ready.

  Try: “Show me the results from the scrape that just finished.”

  Network · get_scraper_results

• Reloads the scraper catalog

  Refreshes the list of available scrapers right away instead of waiting for the daily update.

  Try: “Refresh the scraper list.”

  Network · refresh_scrapers

• Opens a page in a remote browser

  Starts or reuses a remote browser session and goes to the URL you give it.

  Try: “Open the login page of this site in a browser.”

  Network · scraping_browser_navigate

• Lists the clickable parts of the open page

  Returns an accessibility snapshot of the page with a reference for each button, link and field. Other browser actions use these references, so take a new one after the page changes.

  Try: “Show me what buttons and fields are on this page.”

  Network · scraping_browser_snapshot

• Clicks something on the page

  Clicks the element you point to from the latest snapshot. This can submit forms or trigger actions on the site.

  Try: “Click the "Load more" button.”

  Write · scraping_browser_click_ref

• Types into a field on the page

  Types text into a field from the latest snapshot and can press Enter to submit it.

  Try: “Type "wireless headphones" in the search box and press Enter.”

  Write · scraping_browser_type_ref

• Fills in several form fields at once

  Fills multiple fields of a form in one call, using references from the latest snapshot.

  Try: “Fill in the contact form with my name and email.”

  Write · scraping_browser_fill_form

• Takes a screenshot of the page

  Captures an image of the current page, or the full page if you ask for it.

  Try: “Take a full-page screenshot of this page.”

  Network · scraping_browser_screenshot

• Reads the text of the open page

  Returns the text content of the page body in the remote browser.

  Try: “Read the text on the page after it finished loading.”

  Network · scraping_browser_get_text

• Gets the HTML of the open page

  Returns the HTML of the page currently open in the remote browser.

  Try: “Get the HTML of the page as it looks now.”

  Network · scraping_browser_get_html

• Goes back one page

  Moves the remote browser back to the previous page.

  Try: “Go back to the results list.”

  Network · scraping_browser_go_back

• Goes forward one page

  Moves the remote browser forward in its history.

  Try: “Go forward again.”

  Network · scraping_browser_go_forward

• Scrolls to the bottom of the page

  Scrolls the open page to the end, which loads content on pages that fill in as you scroll.

  Try: “Scroll to the bottom so all items load.”

  Network · scraping_browser_scroll

• Scrolls a specific element into view

  Scrolls the page until the element you choose from the snapshot is visible.

  Try: “Scroll to the pricing table.”

  Network · scraping_browser_scroll_to_ref

• Waits for something to appear

  Waits until an element shows up on the page, with an optional time limit.

  Try: “Wait until the results table appears.”

  Network · scraping_browser_wait_for_ref

• Lists the requests the page made

  Returns the network requests since the page loaded, with method, URL and status.

  Try: “Which API calls did this page make?”

  Network · scraping_browser_network_requests

• Asks ChatGPT a question and returns its answer

  Sends your prompt to ChatGPT through Bright Data and returns the answer as markdown. Useful for checking how ChatGPT describes a brand.

  Try: “Ask ChatGPT what the best proxy provider is and show me the answer.”

  Network · chatgpt_ai_insights

• Asks Perplexity a question and returns its answer

  Sends your prompt to Perplexity through Bright Data and returns the answer as markdown.

  Try: “Ask Perplexity which CRM tools are best for startups.”

  Network · perplexity_ai_insights

• Shows which fields a dataset can be filtered by

  Lists the filterable fields of a stored LinkedIn dataset, with their types. It is called before searching the dataset.

  Try: “What fields can I filter LinkedIn company data by?”

  Network · list_dataset_fields

• Finds records in a LinkedIn dataset

  Searches a stored LinkedIn dataset with a filter you describe and returns up to 10 matching records per call, with a cursor for the next page.

  Try: “Find LinkedIn companies in Berlin with 50 to 200 employees.”

  Network · search_dataset

• Asks Bright Data's support assistant

  Sends a question about Bright Data products, zones, errors or account setup to its assistant, and lets you follow up in the same thread.

  Try: “Why is my scraper zone returning an error?”

  Network · ask_brightdata_assistant

• Shows how many times each tool was used

  Returns the number of calls made to each tool during the current session.

  Try: “How many tool calls have I used this session?”

  Read · session_stats

Watch out for

• The remote browser can click and submit forms

  Three browser tools click, type and fill forms on live sites, and the source marks them as not read-only. Leave out the browser group if you do not need it. github.com

• Usage is billed after the free tier

  Scraper runs are billed per record and browser traffic per GB once you have deposited funds. Set a spend cap in the Bright Data control panel. github.com

• Web pages can try to steer the assistant

  Search and scrape results return whatever a page says. A content gate asks you before returning text that looks like an instruction to change your MCP setup, but it does not cover the browser tools. github.com

• Hosted option puts the token in the web address

  The hosted server at mcp.brightdata.com takes your API token as a ?token= part of its URL. Browsers, proxies, shell history, config files in git and screenshots can all record a URL, and the token is account-wide. Use the local install if you cannot keep the address private, and rotate the token if it leaks. github.com

Security advisories