Bright Data MCP Server

Bright Data MCP Server is the official Model Context Protocol interface from Bright Data. It connects an MCP-capable AI client to Bright Data's web unblocking infrastructure and, according to the provider, exposes 69 tools in eleven groups covering web search, page scraping, structured data extraction from well-known platforms, and browser automation. The server is available as a hosted remote endpoint and as a local npm process; both routes send requests through Bright Data's public unblocking network, which according to the provider automatically handles bot detection, CAPTCHA solving, and proxy rotation for every request.

Tools and categories

Base tools are active in every configuration: search_engine and search_engine_batch return results from Google, Bing, and Yandex as structured JSON. scrape_as_markdown and scrape_batch fetch any public URL as clean Markdown. discover uses AI-relevance ranking to find the best sources for a research question with optional date filters.

Optional tool groups extend coverage. The ecommerce group contains eleven tools for Amazon, Walmart, eBay, Best Buy, Etsy, Home Depot, Zara, and Google Shopping that return products, reviews, and search results as clean JSON. The social group has 23 tools for LinkedIn, Instagram, TikTok, Facebook, YouTube, X, and Reddit. The browser group provides 13 tools for remote browser automation without a local Playwright installation. The business group covers Crunchbase, ZoomInfo, Google Maps Reviews, and Zillow. The geo and research groups let agents collect ChatGPT, Grok, and Perplexity answers as structured data and read files from GitHub repositories. The code group returns npm and PyPI package metadata.

Remote endpoint and local installation

According to Bright Data, the recommended route is the hosted remote endpoint https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN. This endpoint requires no local installation and is documented for Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, Zed, and Warp. Alternatively, the server can run locally with npx @brightdata/mcp, with API_TOKEN set as an environment variable. Both routes send requests through Bright Data's network; actual processing happens on Bright Data's infrastructure, not locally.

Free tier and pricing model

According to the provider, a free usage allowance is available for initial testing and is refreshed regularly. After it is consumed, paid use may be available depending on the account. Bright Data publishes plan and allowance details on its official pricing page; this entry states no amounts because they can change. A Bright Data account and API token are required in all cases.

Tool selection and token consumption

Loading all 69 tools at once means significant token use in the model context. Bright Data recommends activating only the needed groups via the GROUPS parameter. A research task only needs base tools; e-commerce monitoring additionally requires ecommerce; social-media tracking needs the social group. Individual tools can be added on top via the TOOLS parameter. A deliberately small tool set reduces token consumption, shrinks the prompt-injection surface, and makes agent decisions easier to review.

Scraping boundaries, terms of use, and personal data

Web scraping does not remove obligations under website terms of use, robots guidance, copyright, data protection law, or contracts. For every target and every tool call, check whether the intended collection, storage, and reuse are permitted. Social-media, profile, and contact data can contain personal data. Establish purpose, legal basis, retention, access, and deletion in advance; do not put secrets or confidential customer data in tool inputs unless this is genuinely necessary and legally supported.

Prompt injection and security

Scraped web pages, search results, and structured platform data are untrusted third-party content. They can carry prompt-injection text: instructions to ignore rules, extract tokens, or trigger further actions. Treat every returned piece of content as data, never as instructions. Limit the agent to the minimum required tool group, review the inputs and recipients of every request, and independently confirm write or publishing actions. The API token grants access to the Bright Data account; store it only as a secret, never in Git, prompts, screenshots, or shared configuration files.

FAQ

Is the Bright Data MCP Server official? Yes. Bright Data operates the repository at github.com/luminati-io/brightdata-mcp and the remote endpoint mcp.brightdata.com as official products.

Which clients are supported? According to the documentation, Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, Zed, and Warp are explicitly documented. Any MCP-compatible client that supports remote SSE or local stdio processes should work.

Can I test the server without an account? According to the provider, a Bright Data account and API token are always required to connect; current trial and usage terms are available in the official documentation.

On 2026-09-08, the GitHub API reported exactly 2,634 stars for luminati-io/brightdata-mcp. This is a point-in-time repository-popularity signal, not evidence of quality, privacy, or security.

A Bright Data account and API token; for the remote endpoint an MCP client with remote SSE support; for local mode Node.js 18+ and a securely set API_TOKEN.

Remote: add https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN in the client. Local: set API_TOKEN securely and configure npx @brightdata/mcp as the stdio command. Activate only required tool groups via the GROUPS parameter.

npx @brightdata/mcp

API token from Bright Data account settings. Passed as a URL parameter for the remote endpoint, or as the API_TOKEN environment variable for the local server.

The API token acts in the Bright Data account context. Access, quota, and billing limits depend on the account type and configured caps; activate only necessary tool groups.

All tool requests are processed through Bright Data's network. Return values — Markdown content, structured JSON, search and browser results — reach the MCP client and may be processed or retained in model context, chat history, or logs.

Scraped content can involve personal data, website terms, and prompt injection. The API token grants account access and must never be stored in plaintext. Minimize the tool set, review inputs and targets, and independently confirm state-changing actions.