Tavily Search provides web search functionality for OpenClaw agents using the Tavily API, serving as a powerful alternative to built-in search tools. It returns structured results with titles, URLs, snippets, and optional answer summaries in multiple output formats — making it easy to integrate web research into any agent workflow.
Key Features
- Multiple output formats including raw JSON, Brave-compatible schema, and clean Markdown lists for flexible downstream processing
- Optional answer summaries that extract short, direct answers from search results when available
- Configurable result count with recommended 3-5 results per query for optimal context usage
- Lightweight Python implementation that wraps the Tavily search API with minimal dependencies
- Environment variable configuration supporting both system environment variables and OpenClaw's
.env
file for API key storage
Use Cases
- Augmenting agent responses with current web information and source citations
- Researching topics, competitors, or market data as part of analysis workflows
- Finding documentation, tutorials, or code examples to assist with development tasks
- Providing agents with real-time information beyond their training data cutoff
How It Works
The skill executes a Python script that accepts a search query and optional parameters, sends the request to the Tavily search API at
api.tavily.com/search
, and returns structured results. Output can be formatted as raw JSON for programmatic use, Brave-compatible schema for interoperability with tools expecting that format, or clean Markdown lists for human-readable output.
Getting Started
Obtain a Tavily API key and configure it via the
TAVILY_API_KEY
environment variable or in your
~/.openclaw/.env
file. The skill runs from the OpenClaw workspace using Python 3 with no additional installations required. Available under the MIT-0 license with over 70,000 downloads from the ClawHub registry.
