| name | tavily |
| description | Web search, content extraction, site crawling, and deep research via Tavily API. Use this skill when the user needs to search the web, extract content from URLs, crawl websites, map site structure, or do deep research on a topic. Trigger on: web search, find online, extract page, crawl site, research topic, "what's happening with X", site map, scrape URL, or any request for current web information. Prefer this over built-in WebSearch/WebFetch when you need advanced features: domain filtering, date ranges, image extraction, deep research, or site crawling.
|
Tavily CLI
CLI for web search, extraction, crawling, and research. Runs through MCPHub (localhost:9700).
tavily <subcommand> [flags]
All commands support -o text|json|markdown|raw (default: text).
Interactive Workflow
Choose the right tool based on what the user needs:
Quick web search
For current information, news, facts:
tavily tavily-tavily-search --query "kubernetes 1.32 new features"
tavily tavily-tavily-search --query "anthropic claude" --time-range week
tavily tavily-tavily-search --query "react server components" \
--raw '{"include_domains": ["react.dev", "github.com"]}'
tavily tavily-tavily-search --query "MCP protocol specification" \
--search-depth advanced --max-results 10
tavily tavily-tavily-search --query "bitcoin price" --search-depth ultra-fast
Present results with titles, snippets, and source URLs. Ask the user if they want to extract full content from any result.
Extract content from URLs
When user provides a URL or wants full page content:
tavily tavily-tavily-extract --urls "https://example.com/article"
tavily tavily-tavily-extract --urls "https://url1.com,https://url2.com"
tavily tavily-tavily-extract --urls "https://linkedin.com/in/someone" \
--extract-depth advanced
tavily tavily-tavily-extract --urls "https://example.com" --include-images true
Crawl a website
When user needs content from multiple pages on a site:
tavily tavily-tavily-crawl --url "https://docs.example.com" \
--max-depth 2 --limit 10
tavily tavily-tavily-crawl --url "https://docs.example.com" \
--instructions "only return API reference pages" --limit 20
tavily tavily-tavily-crawl --url "https://example.com" \
--raw '{"select_paths": ["/docs/.*", "/api/.*"]}'
Map site structure
To see what pages exist without extracting content:
tavily tavily-tavily-map --url "https://docs.example.com" \
--max-depth 3 --limit 50
Deep research
For comprehensive multi-source research on complex topics:
tavily tavily-tavily-research \
--input "Compare Kubernetes vs Nomad for container orchestration in 2026" --model mini
tavily tavily-tavily-research \
--input "State of WebAssembly adoption in production systems" --model pro
Commands Reference
| Command | Purpose | Key Flags |
|---|
tavily-tavily-search | Web search | --query TEXT --search-depth basic|advanced|fast|ultra-fast --time-range day|week|month|year --max-results N |
tavily-tavily-extract | Extract page content | --urls URL1,URL2 --extract-depth basic|advanced --format markdown|text |
tavily-tavily-crawl | Crawl website pages | --url URL --max-depth N --max-breadth N --limit N --instructions TEXT |
tavily-tavily-map | Map site URL structure | --url URL --max-depth N --limit N |
tavily-tavily-research | Deep multi-source research | --input TEXT --model mini|pro|auto |
When to use Tavily vs built-in tools
| Need | Use |
|---|
| Quick factual search | Built-in WebSearch (faster) |
| Domain-filtered or date-filtered search | Tavily search |
| Extract full page content | Tavily extract (better than WebFetch for protected sites) |
| Crawl multiple pages | Tavily crawl (no built-in equivalent) |
| Map site structure | Tavily map (no built-in equivalent) |
| Deep multi-source research | Tavily research (no built-in equivalent) |
Tips
- Use
--raw '{"key": "value"}' for complex parameters like include_domains, exclude_domains, select_paths
--search-depth ultra-fast is great for quick factual lookups
--extract-depth advanced handles LinkedIn, paywalls, and embedded content better
tavily-research --model pro takes longer but covers more sources — use for broad topics
- MCPHub must be running:
make status in /home/danil/code/mcp-hub