Brightdata Mcp— agent skill

<div align="center"> <a href="https://brightdata.com/ai/mcp-server"> <img src="https://github.com/user-attachments/assets/c21b3f7b-7ff1-40c3-b3d8-66706913d62f" alt="Bright Data Logo"> </a>

<h1>Bright Data MCP</h1>

<p><strong>Web search, page scraping, structured data extraction, and browser automation for AI agents and LLMs over the Model Context Protocol.</strong></p> <p>Works with AI agents, coding agents, chat assistants, and any MCP-compatible client.</p>

<p> <a href="https://www.npmjs.com/package/@brightdata/mcp"><img src="https://img.shields.io/npm/v/@brightdata/mcp?style=flat-square" alt="npm version"></a> <a href="https://www.npmjs.com/package/@brightdata/mcp"><img src="https://img.shields.io/npm/dw/@brightdata/mcp?style=flat-square" alt="npm downloads"></a> <a href="https://github.com/brightdata-com/brightdata-mcp/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue?style=flat-square" alt="License"></a> </p>

<p> <a href="#quick-start">Quick Start</a> • <a href="#pricing-and-free-tier">Pricing</a> • <a href="#use-cases">Use Cases</a> • <a href="#tools-reference-32-tools">Tools</a> • <a href="#agent-skills">Agent Skills</a> • <a href="#documentation">Docs</a> • <a href="#support">Support</a> </p>

<p><strong>Free tier: 5,000 requests per month.</strong> No credit card required. Renews monthly.</p>

</div>

Overview

The Bright Data MCP server gives AI agents real-time access to public web data. It exposes 32 tools covering:

• Web search — Bing and Google results as structured data
• Page scraping — any URL as Markdown or HTML, with bot detection, CAPTCHA solving, and proxy rotation handled automatically on every request
• Structured data (Web Scraper API) — clean JSON from 1,200+ pre-built scrapers: Amazon, LinkedIn, Instagram, TikTok, YouTube, X, Reddit, Facebook, Crunchbase, Zillow, npm, PyPI and more, without parsing HTML. Six tools find and run any of them
• Browser automation — navigate, click, type, screenshot, and read pages in a remote browser session
• LLM response collection — send prompts to ChatGPT and Perplexity and get their answers back as structured data

Every request is routed through Bright Data's unblocking infrastructure, so pages that block ordinary HTTP clients (bot detection, CAPTCHAs, rate limits, geo-restrictions) return normally. No proxy setup, no headless browser maintenance, no retry logic to write.

Two deployment options: a hosted remote server (one URL, no installation) or a local instance via npx @brightdata/mcp.

Quick Start

Hosted server — no installation. Add this URL to your MCP client:

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

Get your API token from your Bright Data account settings. New accounts get 5,000 free requests per month.

Optional URL parameters:

<details> <summary><b>Claude Desktop</b></summary>

1. Go to: Settings → Connectors → Add custom connector
2. Name: Bright Data
3. URL: https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN
4. Click "Add"

Or run locally:

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

</details>

<details> <summary><b>Claude Code</b></summary>

claude mcp add --transport http brightdata "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"

</details>

<details> <summary><b>Cursor</b></summary>

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "brightdata": {
      "url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
    }
  }
}

</details>

<details> <summary><b>VS Code</b></summary>

Add to .vscode/mcp.json:

{
  "servers": {
    "brightdata": {
      "type": "http",
      "url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
    }
  }
}

</details>

<details> <summary><b>Windsurf</b></summary>

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "brightdata": {
      "serverUrl": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
    }
  }
}

</details>

<details> <summary><b>Gemini CLI</b></summary>

Add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "brightdata": {
      "httpUrl": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
    }
  }
}

</details>

<details> <summary><b>Zed</b></summary>

Add to your Zed settings:

{
  "context_servers": {
    "brightdata": {
      "url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
    }
  }
}

</details>

<details> <summary><b>Warp</b></summary>

Go to Settings > MCP Servers > Add MCP Server and add:

{
  "brightdata": {
    "url": "https://mcp.brightdata.com/mcp?token=YOUR_API_TOKEN"
  }
}

</details>

<details> <summary><b>Other clients (local npx)</b></summary>

For any client that supports local MCP servers:

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

</details>

Pricing and Free Tier

Every account includes a recurring monthly free tier. No credit card or commitment required to start.

5,000 free requests per month, renewing on the 1st of each month. Unused requests don't roll over. For team accounts, the free tier is shared across all users in the account.

What's included free:

• Fetch any webpage and extract as Markdown
• Access to 60+ pre-built scrapers for popular domains
• Web search (Bing, Google)
• Web unlocking (bot detection bypass, CAPTCHA solving, proxy rotation)
• Browser automation
• Geo-targeting

Beyond the free tier — pay as you go, no commitment:

• When free requests run out, requests stop. No surprise charges — unless you have deposited funds
• Adding a credit card is a verification step only; you are not charged unless your free tier is exhausted and you have funds deposited
• Set a spend cap in the control panel so pay-as-you-go usage never exceeds your budget

Full pricing, volume plans and enterprise →

Use Cases

Real-time research

Answer questions using live web data instead of training data. Search, then read the sources.

Example prompts: "What's Tesla's current stock price?", "Get today's weather forecast for New York", "Find the most cited sources on EU AI regulation from the last 6 months".

E-commerce intelligence

Read product data as structured JSON: price, availability, rating, review count, seller, images.

Example prompts: "Compare this laptop's price on Amazon vs Walmart vs Best Buy", "Get the rating and review count for ASIN B0D2Q9397Y", "Is this product in stock?".

Market and competitor analysis

Build competitor profiles from live data: funding, headcount, hiring, customer reviews, pricing pages.

Example prompt: "Analyze Notion as a competitor: pricing, funding, hiring focus, and what customers complain about".

AI agents with reliable web access

Replace built-in fetch/search tools that get blocked on protected sites. Every request goes through unblocking infrastructure, so agents don't fail on bot detection, CAPTCHAs, or geo-restrictions.

Coding agents

Package registry data on demand — no scraping, no stale caches.

Example prompts: "What's the latest version of express on npm?", "Get the README for the langchain-brightdata PyPI package".

GEO and brand visibility

Send prompts to major LLMs and get their answers back as structured data. Measure how AI assistants describe your brand, which sources they cite, and what they recommend — the feedback loop for Generative Engine Optimization.

Example prompt: "Ask ChatGPT and Perplexity 'what is the best proxy provider' and compare how each one ranks us".

Social media monitoring

Structured data from every major platform: profiles, posts, comments, engagement metrics. Each platform has scrapers per data type (LinkedIn profiles, Instagram reels, TikTok comments, ...); find them with search_scrapers and run them with run_scraper.

Example prompt: "Get the last 10 posts from this TikTok profile and summarize the engagement".

Content creation and academic research

Gather source material from many pages at once, filtered by recency and relevance.

How It Compares

Tool Selection: Groups

Tools are organized into groups so you only load what you need. Fewer tools means less context for your agent to process.

• With no GROUPS or TOOLS, all 32 tools are enabled. Setting either one limits the server to those tools only
• GROUPS enables tool bundles. Comma-separated: GROUPS="scrapers,browser" (local) or &groups=scrapers,browser (hosted URL)
• TOOLS adds individual tools on top: TOOLS="extract,scrape_as_html". On its own, it enables only the tools you list
• Every group also includes search_engine, scrape_as_markdown and ask_brightdata_assistant
• Group ID custom is reserved; use TOOLS for individual picks

Older group IDs (ecommerce, social, business, finance, research, app_stores, travel, code) still work and enable the scrapers tools; social and business also add search_dataset and list_dataset_fields. Old web_data_* names in TOOLS enable the scrapers tools too.

Configuration examples

Local server with browser automation and AI extraction:

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

Coding agent setup (Claude Code / Cursor / Windsurf) — npm, PyPI and GitHub data through the pre-built scrapers:

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

Tools Reference (32 Tools)

Which tool to use

• Known URL, need the content: scrape_as_markdown. Multiple URLs (up to 5): scrape_batch
• Need to find information: search_engine. Multiple queries (up to 5): search_engine_batch
• Structured data from a known platform (Amazon, LinkedIn, TikTok, npm, PyPI, etc.): search_scrapers → get_scraper_details → run_scraper — returns clean JSON, faster and more reliable than scraping the same page
• Structured JSON from a page no scraper covers: extract
• Raw HTML: scrape_as_html
• Page requires interaction (click, type, scroll, login): scraping_browser_* tools
• How ChatGPT/Perplexity answer a prompt: chatgpt_ai_insights / perplexity_ai_insights
• Question about Bright Data itself (products, zones, errors, setup): ask_brightdata_assistant

Notes on scraper runs:

• Return structured JSON, billed per record returned
• Records with an error field explain inputs that failed, e.g. a URL of the wrong type
• Results can be large. Keep limit_per_input low for discover_by_* methods and run bulk collection in a subagent where your framework supports it, so records don't flood the main context window
• If a run fails, scrape_as_markdown works on the same URL as a fallback

<details> <summary><b>Search and Scraping — 7 tools</b></summary>

</details>

<details> <summary><b>Web Scraper API — 6 tools</b></summary>

Search and run any of Bright Data's 1,200+ pre-built scrapers. The scraper list is loaded from docs.brightdata.com/scrapers.json and refreshed daily. Available on both the hosted and the local server. Enabled by default when no tools or groups are selected; otherwise add the scrapers group (&groups=scrapers on the hosted URL, GROUPS=scrapers locally).

Typical flow: search_scrapers → get_scraper_details → run_scraper. Runs are billed per record; discover_by_* methods collect up to limit_per_input records per input (default 10).

</details>

<details> <summary><b>Browser Automation — 14 tools</b></summary>

Remote browser session. Typical sequence: navigate → snapshot → interact by ref → extract or screenshot.

Refs come from the latest snapshot. If the page changes after a click or navigation, take a new snapshot before the next ref-based action. For static pages, scrape_as_markdown is faster and cheaper than a browser session.

</details>

<details> <summary><b>GEO and LLM Visibility — 2 tools</b></summary>

Use for Generative Engine Optimization (tracking how LLMs describe your brand) and LLM-as-a-judge workflows.

</details>

<details> <summary><b>Dataset Search — 2 tools</b></summary>

Filter search over stored LinkedIn datasets; returns matching records at once, with no scraper run.

</details>

<details> <summary><b>Bright Data Assistant — 1 tool</b></summary>

</details>

Full tool reference in the docs →

Agent Skills

Ready-to-use skills that teach your agent how to use this MCP server correctly. The full collection lives at github.com/brightdata/skills — 21 skills covering MCP orchestration, competitive intelligence, price comparison, brand listening, SEO audits, scraper building, RAG pipelines, and more.

Three of the highest-impact skills:

Browse all 21 skills →

Configuration

Basic setup (local)

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

Advanced configuration

{
  "mcpServers": {
    "Bright Data": {
      "command": "npx",
      "args": ["@brightdata/mcp"],
      "env": {
        "API_TOKEN": "your-token-here",
        "RATE_LIMIT": "100/1h",
        "SERP_ZONE": "custom_serp",
        "WEB_UNLOCKER_ZONE": "custom",
        "BROWSER_ZONE": "custom_browser"
      }
    }
  }
}

Environment variables

Content gate

scrape_as_markdown, search_engine, extract and their batch variants return whatever a web page says. If that content contains text that reads like instructions to change your MCP configuration — register a server, run an npx -y package, edit the MCP config file — the server pauses and asks you, through your MCP client, before returning it. Approve and the content is returned unchanged; decline and it is withheld with a short explanation to the assistant. Pages that do not contain such text are returned exactly as before, with no prompt. For a batch call, only the matching items are withheld.

Pages that are about MCP configuration (documentation, READMEs) will naturally prompt. Clients that do not support MCP elicitation get the matching content withheld; set CONTENT_GATE=off if you accept the risk. While the server waits for your answer it reports progress; clients that do not extend their tool timeout on progress will cap the wait at their own limit — CONTENT_GATE_TIMEOUT is the knob on the server side. Browser tools are not covered.

Documentation

Troubleshooting

<details> <summary><b>Common issues and solutions</b></summary>

"spawn npx ENOENT" error

Install Node.js, or use the full path to node:

"command": "/usr/local/bin/node"  // macOS/Linux
"command": "C:\\Program Files\\nodejs\\node.exe"  // Windows

Timeouts on complex sites

Increase the timeout in your client settings to 180s.

Authentication issues

Verify your API token is valid and has the required permissions. Tokens are managed in account settings.

run_scraper returns errors or no data

Records with an error field explain inputs that failed. Compare your input with sample_input from get_scraper_details (e.g., Amazon product URLs contain /dp/, Walmart ones /ip/) and check the page is publicly accessible. scrape_as_markdown works on the same URL as a fallback.

Remote server connection fails

Check your internet connection and firewall settings.

</details>

Contributing

• Report bugs
• Suggest features
• Submit PRs

Please follow Bright Data's coding standards.

Support

License

MIT © Bright Data Ltd.