Tavily
Tavily describes itself as the web layer for AI agents. One API handles real-time web search, page content extraction and site crawling. It returns clean content that a model can read directly.
What Tavily offers
Credit rules and features come from the Tavily website, its pricing page and its credit guide, read on October 24, 2026. Research is billed dynamically per task, so the Tavily console shows the real usage.
| Tavily capability | What Tavily states | Credits used |
|---|---|---|
| Search | Real-time web search at a basic or advanced depth | 1 credit for basic, 2 for advanced |
| Extract | Page content from URLs you provide | Per 5 successful URLs: 1 credit basic, 2 advanced; failed extractions are free |
| Map | Lists the pages of a site | 1 credit per 10 pages returned, 2 with natural-language instructions |
| Crawl | Mapping and extraction combined | The sum of both |
| Research | Research reports with citations | Dynamic, with minimum and maximum per request |
The Tavily website also states:
- The free plan has 1,000 credits per month and needs no card. Pay-as-you-go costs $0.008 per credit. Monthly plans cost $0.0075 to $0.005 per credit.
- Free credits reset on the first day of each month. When they run out, requests stop until the reset or an upgrade.
- Requests pass through security, privacy and content checks that block personal data leaks, prompt injection and malicious sources.
- You can connect Tavily through MCP to Claude, Cursor and other tools. Python and JavaScript SDKs are also documented.
- Startups can apply for 100,000 free credits, and students use it free of charge.
Who it is for
- AI agent developers whose models need live web pages and who want to write less crawling and cleanup code.
- Teams building retrieval-based Q&A that need results with sources, so answers can be traced.
- Analysts doing research and due diligence who want an agent to collect and summarize material on a topic.
An influencer research agent: Tavily for context, EveryInfra for creators
A brand planning creator campaigns in China has two sets of questions up front: what is happening in the category right now, and which creators fit and what they charge. One agent can connect both.
- Context. Use Tavily Search for category news and competitor moves, then Tavily Extract to read a few key articles in full, and let the model write a short market brief. Tavily screens out prompt injection and malicious sources, which protects an agent that feeds web pages straight to a model.
- Creators. Use the EveryInfra Douyin creator search, Xingtu creator analytics and rate cards and Xiaohongshu user search endpoints to list candidates by category keyword, with audience profile, recent post performance and published rates.
- Recommend. Let the model combine both parts into a shortlist that shows each creator's followers, engagement rate, rate card and link to the category news.
In frameworks such as LangChain, Tavily and EveryInfra can both be registered as tools for the same agent. Teams on Claude or Cursor can add both MCP servers; the EveryInfra side is covered in the MCP setup guide. EveryInfra also has its own web search and reading tools (EverySearch), so the two overlap for general search; test the same queries on each.
Start with Tavily's free credits
The Tavily free plan includes 1,000 credits a month with no card, and a basic search costs 1 credit, enough to run this research flow several times. Startups can apply for 100,000 free credits. Tavily can change credit costs per endpoint; tavily.com has the current rules.