Tavily CLI
The Tavily CLI gives agents a native way to access the web from the terminal. Run search, extract, crawl, and research as simple commands with structured output designed for agent workflows.

Why CLI?
Agents already live in the terminal, from OpenClaw to Claude Code to Codex. They plan, write code, execute and iterate through commands. When they need web data they should run a command. The Tavily CLI makes that possible, with every command returning optional structured --json output that agents can parse directly.
Install in one line
Get started directly from your terminal:
curl -fsSL https://cli.tavily.com/install.sh | bash
tvly loginRun your first queries:
tvly search "latest developments in AI agents" --json
tvly extract “https://docs.openclaw.ai/tools/tavily” --json
tvly crawl “https://docs.tavily.com/welcome” --objective "find documentation on n8n integration" --json
tvly research "competitive landscape of AI code assistants" --jsonWhat your agent can do
Search the web in natural language
Agents do not need to construct keyword queries. They can express intent directly and get results that are already shaped for reasoning.
tvly search "Series B AI infrastructure companies founded after 2023" --json
Results come back ranked, filtered, and compressed for LLM use. No raw HTML, no scraping step, no cleanup required before the data is usable.
Use search as part of a reasoning loop
Search is not a one-off call. Agents can use it iteratively.
They can:
- Generate a query
- Inspect results
- Refine the query
- Search again
All within the same task.
How agents actually use it
A typical loop looks like this:
- The agent identifies a gap in its knowledge
- It runs a Tavily search command
- It parses the structured results
- It continues reasoning with fresh information
That simplicity matters because agents are sensitive to friction. The fewer steps between “need information” and “get information,” the more reliable the system becomes.
Get started
Install the CLI, add your API key, and give your agent something that requires real-world knowledge.
curl -fsSL https://cli.tavily.com/install.sh | bash
tvly login
tvly search "your first query" --jsonTry a simple query first.
Then try something harder. Ask for comparisons. Ask for recent developments. Ask for synthesis across sources.
Watch how the agent interacts with the web.
That’s where it clicks.