Overview

The BrightData MCP Server implements the Model Context Protocol (MCP) to give AI agents live, reliable access to the public web via a single, unified interface. It lets LLM-powered assistants and tools browse, fetch, and extract up-to-date web content without common scraping issues such as IP blocks, CAPTCHAs, or aggressive rate limiting.

Built to integrate with any MCP-compatible client or LLM connector, the server exposes tool groups (e.g., web browsing, package registries, LLM queries) so agents can request structured data or raw pages in real time. There is a hosted option for zero‑setup usage and an open-source package you can run locally or in your infrastructure.

Features

Works with any MCP-compatible LLM or agent (Claude, GPT-based agents, Gemini, Llama integrations).
Hosted server plus local/server deployment via the @brightdata/mcp package.
Enterprise-grade unblocking and anti-bot handling to reduce CAPTCHAs and blocks.
Tool groups for task-specific data (Code group for npm/PyPI, GEO group for LLM brand visibility).
Free tier for prototyping (5,000 requests/month).
Zero-config defaults but configurable via environment variables (groups, token, etc.).
Returns structured results suitable for agent tool use (metadata, readme text, dependency lists).

Installation / Configuration

Hosted (no installation)

Use the hosted endpoint directly in any MCP client by appending your API token.

Example (MCP connector URL):

https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN_HERE

Local / CLI (run via npx)

Run the MCP server locally (useful for testing, self-hosting or CI).

Example mcpServers JSON used by some MCP clients:

{
  "mcpServers": {
    "Bright Data": {
      "command": "npx",
      "args": ["@brightdata/mcp"],
      "env": {
        "API_TOKEN": "<your-api-token-here>"
      }
    }
  }
}

Run directly from shell:

# Run with a token in environment
API_TOKEN="your_token_here" npx @brightdata/mcp

Enable specific tool groups (example: code metadata tools):

# Only enable the 'code' tool group: npm + PyPI helpers
GROUPS="code" API_TOKEN="your_token" npx @brightdata/mcp

Notes

API_TOKEN is required for using BrightData’s hosted server or for running with BrightData backends.
GROUPS is a comma-separated list controlling which tool groups are active; use it to limit surface area for agents.

Available Tools / Resources

BrightData MCP exposes multiple tool groups tailored to common agent needs:

code
- npm: package versions, README content, metadata and dependency lists.
- PyPI: project info, release versions, and package metadata.
geo
- LLM brand visibility tools that query services like ChatGPT, Grok, Perplexity and return structured summaries and citations.
web (general browsing)
- Headless browser fetches, HTML content, screenshots, and extracted metadata suitable for processing by agents.
custom / extensible
- You can enable/disable groups and combine them to meet agent requirements without unnecessary capabilities.

Check the repository for up-to-date list of tool groups and supported endpoints: https://github.com/luminati-io/brightdata-mcp

Use Cases

Real-time research: Ask an agent to fetch the latest news, product pages, or stock-related data and get current, crawlable content rather than stale knowledge.
Coding assistants: Let a coding agent fetch an npm package README or PyPI metadata on demand (versions, dependencies) to provide accurate suggestions or migration guidance.
E‑commerce monitoring: Track product availability and price changes across multiple retailers, with anti-blocking to maintain continuity.
GEO & Brand Visibility: Run periodic checks of how major LLMs mention your product and capture recommendations, citations and sentiment for product/SEO optimization.
Hybrid workflows: Combine tool results (e.g., fetch product page + query an LLM for summarization and comparison) to build higher-level automations.

Example flow: get package info for an npm package

Agent invokes the MCP Code tool (group “code”) for package “express”.
MCP returns structured metadata: latest version, README snippets, dependency list.
Agent uses that data to recommend updates, patching steps, or compare versions.

Quick Tips

For experimentation, add the hosted URL into your LLM client (many desktop/agent UIs support adding a custom MCP connector URL).
Narrow GROUPS to just what your agent needs to reduce risk and speed up responses.
Monitor usage if you are on the free tier (5,000 requests/month) — upgrade or move to self-hosting for higher volume.

Where to find the code and docs

GitHub: https://github.com/luminati-io/brightdata-mcp
Project home / docs and hosted service: https://brightdata.com/ai/mcp-server

For developers new to MCP: start with the hosted connector for a few test queries, then move to local deployment when you need tighter control or higher throughput.

BrightData MCP Server for Web Data Extraction