tavily-search-free

Tavily Search Skill (MCP-based)

This skill utilizes the Tavily MCP server, providing clean, real-time web search results optimized for LLMs and RAG pipelines.

Execution Environment

Path Type	Path	Base Directory
Skill Directory	~/.pi/agent/skills/tavily-search-free/	Fixed Location
Main Script	~/.pi/agent/skills/tavily-search-free/executor.py	Skill Directory
Usage	pi Automatic invocation or manual execution	No manual execution required

API Key Configuration

Supports multiple Key rotation to improve request concurrency.

Configuration Method: In the .env file, separate multiple Keys with commas:

TAVILY_API_KEY=key1,key2,key3

View Key Status：

uv run executor.py --key-status

Available Tools

Tool	Description
tavily_search	Web search (news, facts, data)
tavily_extract	Extract content from URL (markdown/text)
tavily_crawl	Crawl website (configurable depth)
tavily_map	Map website structure
tavily_research	Comprehensive research (multiple sources)

Usage

Method 1: Automatic invocation via pi (recommended)

pi Will automatically invoke this skill for web search, no need to manually execute commands.

Method 2: Manual execution

# 从技能目录执行
cd ~/.pi/agent/skills/tavily-search-free

# 列出所有工具
uv run executor.py --list

# 搜索
uv run executor.py --call '{"tool": "tavily_search", "arguments": {"query": "搜索内容"}}'

# 提取 URL 内容
uv run executor.py --call '{"tool": "tavily_extract", "arguments": {"urls": ["https://example.com"]}}'

# 爬取网站
uv run executor.py --call '{"tool": "tavily_crawl", "arguments": {"url": "https://example.com", "max_depth": 2}}'

# 映射网站
uv run executor.py --call '{"tool": "tavily_map", "arguments": {"url": "https://example.com"}}'

# 综合研究
uv run executor.py --call '{"tool": "tavily_research", "arguments": {"input": "研究主题"}}'

Parameter Description

tavily_search

Parameter	Required	Default Value	Description
query	Yes	-	Search query content
max_results	No	5	Maximum number of returned results
search_depth	No	basic	Search depth:basic/advanced/fast/ultra-fast
time_range	No	null	Time range:day/week/month/year

tavily_extract

Parameter	Required	Default Value	Description
urls	Yes	-	URL list
extract_depth	No	basic	Extraction depth:basic/advanced
format	No	markdown	Output format:markdown/text

Output Format

The script outputs in JSON format, containing search results or extracted content.

Monitoring and Statistics

# 查看状态
uv run executor.py --status

# 查看统计
uv run executor.py --stats

# 查看日志
uv run executor.py --logs 100

Path Description

• Script Location：~/.pi/agent/skills/tavily-search-free/executor.py
• Configuration Location：~/.pi/agent/skills/tavily-search-free/mcp-config.json
• Environment Variables：~/.pi/agent/skills/tavily-search-free/.env
• Dependency Installation: Use uv sync for automatic management

---

Based on the Tavily MCP server, supports 5 tools, progressive loading saves context