Bright Data

Exposes 200+ web scraping and data extraction tools through the Model Context Protocol (MCP) standard, allowing AI agents and LLMs to discover and invoke scraping capabilities via a unified tool registry. Built on FastMCP framework, the server implements tool registration, schema validation (Zod), and request routing to Bright Data's backend infrastructure, enabling seamless integration with MCP-compatible clients (Claude Desktop, Cursor, Windsurf) through stdio transport without custom client implementations.

Automatically handles anti-bot detection, CAPTCHA bypass, and geographic restrictions by routing requests through Bright Data's Web Unlocker API, which manages proxy rotation, header spoofing, and JavaScript rendering transparently. The MCP server abstracts this complexity — agents invoke scraping tools without configuring proxies or handling detection logic; the backend automatically applies anti-detection strategies based on target domain fingerprinting and request patterns.

Implements a modular architecture separating concerns into specialized tool modules (browser_tools.js, web_data_tools.js, general_scraping_tools.js), each handling a category of functionality. The central server.js orchestrator routes requests to appropriate modules, which implement tool-specific logic and return results. This modularity enables independent development, testing, and maintenance of tool categories, and allows selective tool loading based on configuration (e.g., disable browser tools if not needed).

Enables AI agents to control headless Chrome browsers remotely through the Chrome DevTools Protocol (CDP), supporting session management, JavaScript execution, DOM interaction, and screenshot capture. The browser_tools.js subsystem manages browser lifecycle (launch, navigation, interaction), maintains session state across multiple tool invocations, and translates agent commands into CDP protocol messages, allowing agents to automate complex multi-step browser workflows without managing browser processes directly.

Provides 196+ dataset-specific scraping tools tailored to popular platforms (Amazon, LinkedIn, Google Maps, eBay, etc.), each implementing platform-specific parsing logic, pagination handling, and data normalization. Rather than generic HTML scraping, these tools understand platform structure and return normalized, structured data (products, profiles, reviews) with consistent schemas. The MCP server exposes each as a distinct tool with platform-specific parameters, allowing agents to extract data from major platforms without writing custom parsers.

Integrates search capabilities across multiple search engines (Google, Bing, Yandex) through dedicated MCP tools, allowing agents to perform web searches and retrieve ranked results without managing search engine APIs directly. Each search tool handles provider-specific parameters, result parsing, and pagination, returning normalized search results with title, URL, snippet, and ranking metadata. The integration abstracts provider differences, enabling agents to switch search engines or aggregate results across providers.

Implements token-based authentication for Bright Data services through environment variables (API_TOKEN), with optional zone configuration for Web Unlocker (WEB_UNLOCKER_ZONE) and Browser API (BROWSER_ZONE). The server validates tokens at startup and per-request, routing authenticated requests to appropriate Bright Data infrastructure zones. Zone configuration allows teams to use separate quotas, rate limits, and proxy pools for different use cases (e.g., dedicated zone for production scraping vs development testing).

Implements configurable rate limiting through the RATE_LIMIT environment variable (format: limit/time+unit, e.g., '100/1m' for 100 requests per minute), throttling tool invocations to prevent quota exhaustion and API abuse. The server enforces limits at the request level, queuing excess requests and returning rate-limit metadata (remaining quota, reset time) to agents, allowing them to implement backoff strategies or prioritize requests.

Uses stdio (standard input/output) as the transport mechanism for MCP protocol communication, enabling the server to integrate with MCP-compatible clients (Claude Desktop, Cursor, Windsurf) without requiring network configuration or port management. The server reads JSON-RPC 2.0 requests from stdin and writes responses to stdout, allowing clients to spawn the server as a subprocess and communicate through pipes, simplifying deployment and eliminating network security concerns.

Leverages the FastMCP framework to implement automatic tool registration, schema validation, and discovery mechanisms, allowing the server to expose 200+ tools with consistent interfaces. FastMCP handles tool metadata (name, description, parameters), Zod schema validation, and request routing, reducing boilerplate code. Clients can discover available tools via MCP's tools/list endpoint, receiving complete tool metadata including parameter schemas, enabling intelligent tool selection and parameter validation before invocation.

Supports Docker deployment through a containerized server image, allowing teams to deploy the MCP server in isolated environments with environment variable configuration. The Dockerfile packages Node.js, dependencies, and the server code, enabling deployment to Kubernetes, Docker Compose, or cloud container services. Configuration is entirely environment-based (API_TOKEN, RATE_LIMIT, zones), allowing the same image to be deployed across development, staging, and production without code changes.