
research
by tavily-ai
Official Tavily plugin for Claude Code
SKILL.md
name: research description: "Get AI-synthesized research on any topic with citations, directly in your terminal. Supports structured JSON output for pipelines. Use when you need comprehensive research grounded in web data without writing code."
Research Skill
Conduct comprehensive research on any topic with automatic source gathering, analysis, and response generation with citations.
When to Use
- Researching any topic requiring web-sourced information
- Generating structured research reports with custom schemas
- Building pipelines that need AI-synthesized insights with citations
Prerequisites
Tavily API Key Required - Get your key at https://tavily.com
Add to ~/.claude/settings.json:
{
"env": {
"TAVILY_API_KEY": "tvly-your-api-key-here"
}
}
Quick Start
Tip: Research can take several minutes, especially with
--model pro. Press Ctrl+B to run in the background and continue working while it completes.
Basic Research (Polling Mode)
python scripts/research.py "Latest developments in quantum computing"
With Custom Schema
python scripts/research.py "Electric vehicle market analysis" \
--schema ./schemas/market_analysis.json \
--model pro
Streaming Mode
python scripts/research.py "AI agent frameworks comparison" --stream
Save to File
python scripts/research.py "Rust async ecosystem" \
--output ./reports/rust_async.json \
--model pro
CLI Reference
| Option | Short | Default | Description |
|---|---|---|---|
topic | - | Required | Research topic or question |
--schema | -s | None | Path to JSON schema file or inline JSON |
--stream | - | False | Enable streaming mode |
--model | -m | mini | Model: mini, pro, auto |
--citation | -c | numbered | Citation format: numbered, mla, apa, chicago |
--output | -o | stdout | Output file path |
--poll-interval | -p | 5 | Seconds between polls (polling mode) |
--quiet | -q | False | Suppress progress output |
Output Format
{
"meta": {
"topic": "Your research topic",
"model": "pro",
"completed_at": "2025-01-08T14:30:00Z",
"response_time_seconds": 45.2
},
"content": "...",
"sources": [
{"url": "https://...", "title": "Source Title", "citation": "[1]"}
]
}
- content: Markdown string (default) or structured JSON (when schema provided)
- sources: Array of citations used in the research
Schema Usage
Schemas make output structured and predictable. Provide via file path or inline JSON.
File Path
python scripts/research.py "topic" --schema ./my_schema.json
Inline JSON
python scripts/research.py "topic" --schema '{"properties": {"summary": {"type": "string", "description": "Executive summary"}}}'
Schema Requirements
Every property MUST include both type and description:
{
"properties": {
"summary": {
"type": "string",
"description": "2-3 sentence executive summary"
},
"key_points": {
"type": "array",
"description": "Main takeaways",
"items": {"type": "string"}
}
},
"required": ["summary", "key_points"]
}
See references/schema.json for complete schema rules and examples.
Model Selection
Rule of thumb: "what does X do?" → mini. "X vs Y vs Z" or "best way to..." → pro.
| Model | Use Case | Speed |
|---|---|---|
mini | Single topic, targeted research | Fast |
pro | Comprehensive multi-angle analysis, open ended | Slower |
auto | API chooses based on topic complexity | Varies |
Examples
Market Research
python scripts/research.py "Fintech startup landscape 2025" \
--schema ./schemas/market_research.json \
--model pro \
--output ./reports/fintech_2025.json
Technical Comparison
python scripts/research.py "LangGraph vs CrewAI for multi-agent systems" \
--model pro \
--citation mla
Quick Overview
python scripts/research.py "What is retrieval augmented generation?" --quiet
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon