minhngoc25a Tavily MCP
Use the local Tavily MCP server for LLM-optimized web data.
Prerequisites
The Tavily MCP server must already be configured and connected in Claude Code. Do not install or call the Tavily CLI from this skill.
If a Tavily MCP tool is missing, tell the user which local MCP capability is unavailable and continue with local/contextual information only unless they configure the MCP server.
Choose the right tool
Follow this escalation path:
- Search — no specific URL yet; find pages, recent information, or source candidates.
- Extract — one or more known URLs; return clean markdown or text content.
- Map — known site, unknown page; discover URLs without extracting content.
- Crawl — many pages from a site or docs section; extract content in bulk.
- Research — deep, multi-source synthesis or comparison report.
| Need | MCP tool | Skill |
|---|---|---|
| Find pages or quick facts | mcp__tavily-mcp__tavily_search |
tavily-mcp-search |
| Read specific URLs | mcp__tavily-mcp__tavily_extract |
tavily-mcp-extract |
| Discover URLs on a site | mcp__tavily-mcp__tavily_map |
tavily-mcp-map |
| Extract many pages from a site | mcp__tavily-mcp__tavily_crawl |
tavily-mcp-crawl |
| Produce a cited research report | mcp__tavily-mcp__tavily_research |
tavily-mcp-research |
Usage guidance
- Use the most specific skill when the task clearly maps to one tool.
- Prefer
mapplus targetedextractover broadcrawlwhen only a few pages are needed. - Prefer
searchfor quick facts and source discovery. - Prefer
researchwhen the user asks for synthesis, comparison, market analysis, literature review, or a detailed report. - Keep query strings concise and focused; web search queries should be search terms, not long prompts.
- Use domain and path filters when the user names trusted sources or a target site section.