Tavily
Listed at https://tavily.com
Overall Rank #18 ⭐ Consider
⚠️ Tavily API hosted on AWS US-East / EU-West; direct access from mainland China 200-400ms latency; recommended Cloudflare Worker or Tencent Cloud Edge Function proxy (50-100ms); no CN node planned; no domestic payment channels | 🌍 International
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Free 1,000 credits/month(滚动 30 天窗口,无信用卡);Pay-as-you-go $0.008/credit(基础 Search 1 credit);Project 起步 $30/month 含 4,000 credits,$0.008/credit 超额;Growth $200/month 含 40,000 credits,$0.006/credit 超额;Pro $1,000/month 含 250,000 credits,$0.005/credit 超额;Enterprise 合同制(custom 配额 + SLA + 无日志合约) | per million tokens |
| Output | 按 endpoint 计费:Search 1 credit/次;Extract 1 credit/页;Crawl 1 credit/页;Map 1 credit/页;Research 5 credits/次(替代 5-10 Search + 2-3 Extract + 1 合成 LLM);AI Extraction 1 credit/次 | per million tokens |
💡 Free Credits: Free forever: 1,000 credits/month (rolling 30-day window), no credit card, no feature gating; Search/Extract/Crawl/Map/AI Extraction each have separate quotas (1,000 Search + 500 Crawl); overage returns HTTP 429, no surprise billing
🤖 Supported Models (7)
Search (real-time web + AI summary, 1 credit/call)Extract (clean markdown from URL, 1 credit/page)Crawl (BFS domain walk, 1 credit/page)Map (sitemap-level discovery, 1 credit/page)Research (multi-step agent brief, 5 credits/call)AI Extraction (structured JSON schema, 1 credit/call)MCP Server (March 2026, Search + Extract as MCP tools)
✨ Pros
- ✓🎯 Default search API for AI agents: LangChain TavilySearchResults, LlamaIndex TavilyToolSpec, AutoGen/CrewAI/smolagents all default to Tavily
- ✓🆓 1,000 credits/month free tier (no credit card, no feature gating), zero-cost RAG agent prototyping
- ✓🧠 /research endpoint: 5 credits/call, embedded multi-step search + extract + synthesis LLM, single call replaces 50-100 lines of agent code
- ✓🔌 MCP server (launched March 2026, Databricks MCP Marketplace + IBM WatsonX + JetBrains integration) — Claude Code / Cursor / Codex plug-and-play
- ✓💰 Series A $25M funding (2026 H1), Trusted by 2M+ developers, LangChain default-position lock-in
- ✓📦 Search+Extract single-call returns full markdown (answer + raw_content + 0-1 relevance score), eliminates follow-up extraction step, saves 2-5s per research step
⚠️ Cons
- ×Hosted-only, no self-hosting option (vs SearXNG/Meilisearch open-source alternatives)
- ×Direct access from mainland China 200-400ms latency, requires Cloudflare Worker / Edge Function proxy
- ×Research endpoint does not expose internal model stack, cannot audit or switch models (GPT-5.5/Opus 4.5-class but undisclosed)
- ×/research endpoint's answer field occasionally hallucinates on multi-hop queries, requires own-LLM synthesis for safety
- ×Free tier has separate Crawl/Search quotas; a single large Crawl can exhaust the entire monthly Crawl quota
🎯 Best For
Developers building RAG agents / automated research agents (LangChain/LlamaIndex/AutoGen/CrewAI default integration); research workflows needing Search+Extract single-call paradigm; multi-step research agents (budget <$200/month); budget-sensitive solo developers (free tier sufficient for prototyping); zero-integration search inside Claude Code / Cursor / Codex via MCP server