tavily-search skill: have your agent search the web, read pages, map and crawl sites with Tavily

Teaches your agent to search the web, read pages, find a page on a site and crawl many pages, using Tavily.

Installs: 37K; on skills.sh
Stars: 491; on GitHub
Files: 3; besides SKILL.md
Updated: Oct 2026; last repo commit

tavily-search is a skill from Tavily, the search API company. It makes your agent use Tavily for anything that needs the live web, such as current facts, sources, reading a URL or collecting pages from a site. It tells the agent to use advanced search depth, ask for 10 results, run searches in batches and stop once two independent sources agree.

It needs either the Tavily MCP tools or the tvly command-line tool. Search and extract work without an account up to a keyless rate limit, while map, crawl and research need you to sign in. Your queries and URLs go to Tavily's service.

Good for

Researchers who want sourced answers from current web pages
Developers who need an agent to read docs or changelog pages from a site
Anyone whose agent keeps answering from memory instead of checking

Install tavily-search

Claude Code

Copy this prompt.

Install the tavily-search skill in Claude Code. Run this command:

curl -fsSL https://cli.tavily.com/install.sh | bash
tvly init

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

The first route installs the tvly CLI, signs you in, installs the Tavily skills for the agents it finds (Claude Code, Codex, Cursor) and runs a test search. Use tvly init --skip-auth to stay keyless. The second route installs the skills and connects the Tavily MCP server; sign in through the browser when asked. Restart Claude Code after adding the server. Pick one route, not both.

From the project README. Source.

What's in the skill

When it loads

Use this skill whenever a task needs the web: searching for information, checking current facts or news, finding sources, reading a URL, finding a page on a site, or collecting many pages from a site. Use it for anything past your training data or anything the user wants verified online.

Sections in SKILL.md

Search
Extract
When to stop
Hand off the job

Key facts

Author: Tavily (tavily-ai organisation; README is titled Tavily Agent Skills)
Installs on skills.sh: 36,955 (data from 2026-11-02)
GitHub stars: 491 for the whole tavily-ai/skills repo (2026-11-02)
License: MIT
Last repo commit: 2026-10-30
What is inside: 4 files including SKILL.md: references/crawl.md, map.md and setup.md, no scripts
Needs: The tvly CLI or the Tavily MCP tools; sign-in is needed for map, crawl and research

What it includes

tvly search with advanced depth and up to 10 results
Writes search-engine style queries, one focused search per part of a question, and batches independent searches in one turn. Uses time ranges and include_domains filters before raising the result count.
Try: “Search for the Rust 2024 edition release date and confirm it on the official site.”
Runs code · Search
Reads up to 20 URLs per call as markdown
Pulls full pages, including ones built with JavaScript, and retries with advanced extract depth when the basic call comes back empty. It checks failed_results every time.
Try: “Extract the pricing page from these three URLs.”
Runs code · Extract
Lists the URLs on a site without reading them
Used to find the right page on a site before extracting it, or to learn a site's structure before a crawl. Needs sign-in.
Runs code · Map (references/map.md)
Follows links and returns many pages from one site
The guide says to give every crawl instructions and a page limit, start at depth 1 with about 20 pages, and aim at one section. Saving pages as files needs the CLI. Needs sign-in.
Try: “Crawl the /docs section of this site and give me the pages about authentication.”
Runs code · Crawl (references/crawl.md)
Stop at two independent sources, confirm key facts on official sites
For releases, ownership and official figures it wants confirmation on the official domain or repository, not a secondary site.
Read · When to stop
Uses -o to write long output to a file
Long raw page content or crawls are saved with -o and read in parts instead of printed in full.
Write · Save large results to disk
Uses a sibling skill for in-depth reports
For multi-source reports it tells the agent to use tavily-research instead, and to run npx skills add tavily-ai/skills --skill tavily-research if it is missing.
Runs code · Hand off to tavily-research
Connects the MCP server or installs the tvly CLI
Covers claude mcp add for the Tavily MCP server, the tvly install script, tvly init, tvly login and TAVILY_API_KEY.
Runs code · Setup (references/setup.md)

From the official docs

Permissions

Medium risk

The skill sends your searches and URLs to Tavily and its setup guide can run an install script, so a careless prompt could leak private details or use credits. Keep secrets out of queries and approve setup commands yourself.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Your queries and URLs leave your machine
Everything you search, extract or crawl goes to Tavily. Keep private URLs and secrets out of prompts.
Setup can run an install script
references/setup.md uses curl -fsSL https://cli.tavily.com/install.sh | bash, uv tool install tavily-cli or pip install tavily-cli. Install the CLI yourself if you want to review it first.
Crawls can be large and cost credits
The crawl guide says to set a limit and start small because a crawl without one can run much longer and cost much more. Check your plan limits.
Page content comes into the conversation
Fetched pages can contain text aimed at steering the agent. Do not approve actions because a web page asked for them.

Technical details

Talks to	Tavily API and MCP server (mcp.tavily.com). Pages you search, extract, map or crawl
Sign-in and keys	A Tavily account sign-in or TAVILY_API_KEY for map, crawl and research; search and extract can run keyless within a rate limit.
Scopes	Runs tvly commands (allowed-tools: Bash(tvly *)) May save search, extract and crawl output as files Setup guide covers installing the tvly CLI and adding an MCP server

Questions

What does the tavily-search skill do?

It tells your agent when and how to use Tavily for web work. That covers search, extracting page text, mapping a site and crawling pages, with rules for query writing and when to stop.

How do I install it?

Use the install box at the top of this page. You need the skill and also either the tvly command-line tool or the Tavily MCP server, because the skill only gives instructions.

Do I need a Tavily account or API key?

Not to start. Per setup.md, search and extract work without an account up to a keyless rate limit. Map, crawl and research need you to sign in with tvly login, tvly init or a TAVILY_API_KEY.

Is it free?

The skill is MIT licensed and free. Tavily's service has its own limits, and SKILL.md does not give prices. Check Tavily's pricing page for plan details.

Where does my data go?

Search queries and the URLs you extract or crawl are sent to Tavily's service through the tvly CLI or the MCP server. Do not put secrets or private URLs in queries.

What is the difference between tavily-search and tavily-research?

tavily-search handles single facts, page reads and crawls. SKILL.md hands in-depth reports that need many weighed sources to the separate tavily-research skill.

Is it safe?

The skill runs tvly commands and may install the CLI or add the MCP server if it is missing, which you should approve yourself. Crawls can fetch many pages, so set a limit. Treat page text as untrusted and do not let it direct the agent.

Information

updated Nov 2, 2026

Repository: tavily-ai/skills
Folder: skills/core/tavily-search

tavily-ai/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5