bright-data-claude-skill-deep-research（bright 数据 Claude 技能 deep 研究）

Bright Data Research Skill

Advanced web research powered by Bright Data MCP - perform market analysis, competitive intelligence, data extraction, and comprehensive web research with anti-bot protection.

Overview

This skill provides complete workflows for automated web research using Bright Data MCP. Handle search discovery, content collection, structured data extraction, and comprehensive analysis with browser automation support.

When This Skill Applies

Activate this skill when the user's request involves:

• Web scraping and data collection
• Market research and competitive analysis
• Price monitoring and comparison
• Product information extraction
• Search engine result analysis
• Large-scale web data gathering
• Research requiring anti-bot protection

Core Capabilities

Search and Discovery

Use search_engine tool to find relevant sources:

{
  "tool": "search_engine",
  "parameters": {
    "query": "site:etsy.com nba merchandise",
    "engine": "google",
    "cursor": "0"
  }
}

Search strategies:

• Use site operators: "site:etsy.com keywords"
• Use exact phrases: "machine learning in healthcare"
• Exclude terms: "iphone -case -cover"
• Paginate with cursor: "0", "1", "2" for more results

Content Collection

Three collection modes based on research depth:

Quick Mode (3-5 URLs, serial processing):

• Use scrape_as_markdown for each URL
• Best for: Fast overviews, fact-checking

Standard Mode (10-20 URLs, parallel batch):

• Use scrape_batch for up to 10 URLs concurrently
• Best for: Market research, competitive analysis

Deep Mode (20-50 URLs, browser automation):

• Use scraping_browser_navigate for JavaScript-rendered pages
• Use scraping_browser_links to discover page links
• Use scraping_browser_click for interactions
• Best for: Dynamic content, multi-page extraction

Data Extraction

Use extract tool for AI-powered structured data extraction:

{
  "tool": "extract",
  "parameters": {
    "url": "https://example.com/product",
    "extraction_prompt": "Extract: product name, price as number, rating (0-5), number of reviews, seller name, availability status"
  }
}

Common extraction schemas:

• E-commerce: name, price, rating, reviews, seller, availability
• Articles: title, author, date, summary, key points
• Companies: name, industry, founded, headquarters, employee count

Output Formats

Three report formats for different use cases:

Report Format (default):

• Executive summary
• Key findings with evidence
• Detailed analysis
• Methodology and recommendations
• Source references

JSON Format:

• Structured data for API integration
• All raw and processed data
• Metadata and provenance
• Statistical analysis

Markdown Format:

• Clean, readable content
• Tables and lists
• Source links
• Minimal formatting

Research Workflow

Phase 1: Query Analysis

Understand the research intent:

• Scope: How broad/deep should research be?
• Key entities: Products, companies, topics
• Target sources: Which sites/platforms?
• Data needed: What fields to extract?

Phase 2: Source Discovery

Use search_engine to find URLs:

1. Execute initial search
2. Extract URLs from SERP
3. Filter irrelevant domains
4. Paginate if needed
5. Prioritize by relevance

Phase 3: Content Collection

Choose appropriate mode:

• Quick: scrape_as_markdown per URL
• Standard: scrape_batch 10 URLs at once
• Deep: scraping_browser_navigate + browser tools

Handle errors gracefully:

• Retry failed URLs with alternative methods
• Log errors for transparency
• Continue with available data

Phase 4: Data Extraction

Apply extraction schema:

• Use extract with custom prompts
• Validate extracted data
• Handle missing/malformed data
• Ensure data quality

Phase 5: Analysis & Synthesis

Process and analyze:

• Clean and normalize data
• Perform statistical analysis
• Identify patterns and trends
• Cross-reference sources
• Validate findings

Phase 6: Report Generation

Generate output:

• Report: Comprehensive document with all sections
• JSON: Structured data for processing
• Markdown: Clean, readable content

Best Practices

Search Strategy

• Start broad, then narrow down
• Use site operators for targeted searches
• Try multiple search engines if needed
• Set realistic limits (10-20 URLs usually sufficient)

Performance

• Use scrape_batch for parallel processing (10x faster)
• Only use deep mode when necessary (much slower)
• Set appropriate timeouts
• Monitor success rates
• Avoid token limits: Batch 1-2 URLs at a time for large pages (Etsy, Amazon, etc.)

Data Quality

• Always validate extracted data
• Cross-reference multiple sources
• Check for outliers and anomalies
• Normalize formats (dates, currencies, units)

Error Handling

• Implement retry logic
• Have fallback strategies
• Log errors for debugging
• Don't fail on individual URL errors

Ethical Considerations

• Respect robots.txt
• Don't overwhelm servers
• Rate limit requests
• Cite sources properly
• Don't misuse personal data

Common Research Scenarios

E-commerce Market Research

Query: "site:etsy.com nba merchandise"
Mode: standard
Extract: product name, price, rating, reviews, seller
Output: report

Expected: Price analysis, popular products, top sellers

Price Comparison

Query: "iphone 15 pro max 256GB price comparison"
Mode: standard
Extract: retailer, price, availability, shipping
Output: json

Expected: Structured comparison with best deal identified

Academic Research

Query: "machine learning in healthcare 2024 papers"
Mode: standard
Extract: title, authors, date, key findings, methodology
Output: report

Expected: Literature review with trends and insights

Competitive Intelligence

Query: "competitor.com features pricing"
Mode: deep
Extract: feature name, description, pricing tier, availability
Output: report

Expected: Feature comparison, pricing analysis, recommendations

Tool Reference

search_engine

Purpose: Find relevant web pages Parameters: query (required), engine (google/bing/yandex), cursor (page number) Returns: SERP results in markdown

scrape_as_markdown

Purpose: Get clean, AI-ready markdown Parameters: url (required) Returns: Formatted markdown without ads/clutter

scrape_as_html

Purpose: Get raw HTML Parameters: url (required) Returns: Complete HTML document

extract

Purpose: AI-powered structured data extraction Parameters: url (required), extraction_prompt (optional) Returns: JSON object with extracted data

scrape_batch

Purpose: Process multiple URLs in parallel Parameters: urls (array, max 10) Returns: Array of page contents

scraping_browser_navigate

Purpose: Navigate JavaScript-rendered pages Parameters: url (required) Returns: Page info (title, URL, status)

scraping_browser_click

Purpose: Click elements on page Parameters: selector (CSS selector) Returns: Action result

scraping_browser_links

Purpose: Get all links on current page Parameters: None Returns: Array of links with text, href, selector

Troubleshooting

No search results

• Try different search engine (bing, yandex)
• Simplify the query
• Check for typos
• Use broader search terms

Scraping fails

• URL might be JavaScript-rendered → use mode=deep
• URL might be blocked → try alternative URL
• Check if URL is accessible in browser

Extraction incomplete

• Provide more specific extraction prompt
• Check if data exists on page
• Try scraping as markdown first to see content

Slow performance

• Reduce max_results
• Use mode=standard instead of deep
• Check network connectivity
• Close unnecessary browser sessions

Token limit exceeded

• Symptom: "Output exceeds maximum allowed tokens" error
• Cause: Batch scraping too many large pages at once OR reading large files
• Why this limit exists:

- Memory protection: Prevents memory overflow from loading too much content - Performance optimization: Ensures fast response times - Context management: Preserves space for other content in the conversation - System stability: Prevents crashes or errors

• Can this limit be increased?:

- No - This is a hard system limit in Claude Code - Cannot be changed via configuration files - Purpose: Protect system stability and performance

• Workarounds:

- For scraping: Reduce batch size to 1-2 URLs for large pages - For reading files: Use Read with offset and limit to read in chunks - For specific content: Use Grep to search for specific patterns - For finding files: Use Glob to find files by pattern

Additional Resources

Reference Files

For detailed workflows and techniques:

• references/search-discovery.md - Search strategies and URL discovery
• references/content-scraping.md - Content collection methods
• references/data-extraction.md - Extraction schemas and validation
• references/deep-scraping.md - Browser automation techniques
• references/analysis-report.md - Analysis and report generation

Example Files

Complete research examples:

• examples/market-research-etsy-nba.md - E-commerce market research
• examples/competitive-analysis-pricing.md - Price comparison workflow
• examples/academic-research-ml-healthcare.md - Academic literature review

Limitations

• Requires Bright Data MCP server configuration
• Needs valid Bright Data API token
• Subject to API rate limits
• Browser automation is slower than direct scraping
• Some sites may still block access
• Quality depends on source content

Progressive Disclosure

This SKILL.md provides core workflows and quick reference (approximately 2,000 words).

For detailed implementation patterns, advanced techniques, and comprehensive examples, consult the references/ files which load as needed during research tasks.