internet-research

name: internet-research description: Expert internet research using Tavily and Brave Search. Use when the user needs current information beyond knowledge cutoff, real-time data (prices, stats, news), fact verification, documentation lookups, latest package versions, or current events.

Internet Research Skill

This skill provides Claude Code with expert-level internet research capabilities using two complementary MCP servers: Tavily (AI-native search, extraction, crawling) and Brave Search (multimedia search, news, local).

Prerequisites

• Tavily MCP server configured in user scope
• Brave Search MCP server configured in user scope
• TAVILY_API_KEY and BRAVE_API_KEY environment variables set in your shell profile

Setup (if not already done):

# Add Tavily MCP to user scope
claude mcp add tavily --scope user \
  -e TAVILY_API_KEY='${TAVILY_API_KEY}' \
  -- npx -y tavily-mcp@latest

# Add Brave Search MCP to user scope
claude mcp add brave-search --scope user \
  -e BRAVE_API_KEY='${BRAVE_API_KEY}' \
  -- npx -y @brave/brave-search-mcp-server

# Verify both are configured
claude mcp list

When to Use This Skill

Invoke this skill when the user needs:

• Current information beyond your knowledge cutoff
• Real-time data (prices, stats, news)
• Verification of facts or claims
• Documentation lookups
• Latest version information for packages/tools
• Current events or recent developments

Task-Based Tool Routing

These two servers are complementary tools with different jobs, not competing alternatives. Route by task type:

Available Tools — Tavily

mcp__tavily__tavily_search

AI-native web search. Returns pre-cleaned, structured, LLM-optimized results. Best for technical and coding queries.

Parameters:

• query (required): Search terms
• search_depth (optional): "basic" | "advanced" | "fast" | "ultra-fast" (default: "basic")
• max_results (optional): Number of results (default: 5, max: 20)
• time_range (optional): "day" | "week" | "month" | "year"
• include_domains (optional): Array of domains to include
• exclude_domains (optional): Array of domains to exclude
• country (optional): Boost results from specific country

mcp__tavily__tavily_extract

Extract clean content from URLs in markdown or text format.

Parameters:

• urls (required): Array of URLs to extract content from
• extract_depth (optional): "basic" | "advanced" (use advanced for LinkedIn, protected sites)
• format (optional): "markdown" | "text" (default: "markdown")

mcp__tavily__tavily_crawl

Crawl a website with configurable depth and breadth.

Parameters:

• url (required): Root URL to begin crawling
• max_depth (optional): How far from base URL (default: 1)
• limit (optional): Total links to process (default: 50)

mcp__tavily__tavily_map

Map a website's structure, returning discovered URLs.

Parameters:

• url (required): Root URL to begin mapping

mcp__tavily__tavily_research

Comprehensive multi-source research with AI synthesis.

Parameters:

• input (required): Description of the research task
• model (optional): "mini" | "pro" | "auto" (default: "auto")

Available Tools — Brave Search

mcp__brave-search__brave_web_search

General web search. Best for routine queries and preserving Tavily credits.

Parameters:

• query (required): Search terms
• count (optional): Number of results (default: 10, max: 20)
• offset (optional): Pagination offset

mcp__brave-search__brave_image_search

Image search — only available via Brave.

mcp__brave-search__brave_video_search

Video search — only available via Brave.

mcp__brave-search__brave_news_search

News search with freshness controls — only available via Brave.

mcp__brave-search__brave_local_search

Local business/place search — only available via Brave.

Search Strategy

Step 1: Formulate Effective Queries

• Be specific: "Next.js 14 app router migration guide" > "nextjs help"
• Include year for current info: "React best practices 2026"
• Use quotes for exact phrases: "useEffect cleanup function"

Step 2: Route to the Right Tool

Use the task-based routing table above. Key rules:

• Coding/docs/APIs → tavily_search (AI-native quality matters here)
• Images/video/news/local → Brave (only option)
• General web search → brave_web_search (preserves Tavily credits)
• URL content extraction → tavily_extract (cleaner than WebFetch)
• Deep research → tavily_research (built-in multi-source synthesis)

Step 3: Analyze Results

• Check source credibility (official docs > random blogs)
• Verify date freshness
• Cross-reference if critical

Step 4: Synthesize Response

• Summarize key findings
• Cite sources with URLs
• Note any conflicting information
• Indicate confidence level

Example Queries

Response Format

When presenting research results:

## Research Results

**Query:** [what you searched for]

### Key Findings

1. **[Finding 1]**
   - Details...
   - Source: [URL]

2. **[Finding 2]**
   - Details...
   - Source: [URL]

### Summary

[Concise synthesis of findings]

### Sources
- [Source 1](URL)
- [Source 2](URL)

Error Handling

If Tavily fails:

1. Check if TAVILY_API_KEY is set
2. Fall back to mcp__brave-search__brave_web_search
3. Fall back to knowledge cutoff information with disclaimer

If Brave fails:

1. Check if BRAVE_API_KEY is set
2. Fall back to mcp__tavily__tavily_search
3. Fall back to built-in WebSearch

Quota Management

• Route by task to maximize value from both free tiers
• Combined: ~2,000 queries/month with no waste
• Be efficient with queries — cache results mentally within conversation