AgentSkillsCN

research-tools

通过 Context7（文档检索）、Grep.app（代码示例）、Tavily（网页搜索、提取、爬取），以及 Exa（网页搜索）进行外部调研。同时，借助 skill_mcp 功能，按需加载 MCP 数据。

SKILL.md
--- frontmatter
name: research-tools
description: External research via Context7 (docs), Grep.app (code examples), Tavily (web search, extract, crawl), and Exa (web search). Loads MCPs on-demand via skill_mcp.
license: MIT
compatibility: opencode
metadata:
  category: research
  triggers: docs, documentation, code examples, web search, how do others, library, API, current info, crawl, extract, scrape

Research Tools

CRITICAL: skill_mcp Syntax

code
skill_mcp(mcp_name="<MCP_SERVER>", tool_name="<TOOL>", arguments='<JSON>')
  • •mcp_name = MCP server (context7, grep_app, tavily, websearch) — NOT "research-tools"
  • •tool_name = Tool name without prefix — NOT context7_resolve-library-id

Tools

MCP ServerToolUse For
context7resolve-library-idGet library ID (required first)
context7query-docsQuery library documentation
grep_appsearchGitHubGitHub code pattern search
tavilytavily_searchWeb search (real-time, news, finance)
tavilytavily_extractExtract content from specific URLs
tavilytavily_crawlCrawl multiple pages from a base URL
tavilytavily_mapMap/discover website structure and URLs
websearchweb_search_exaWeb search

Examples

Context7 (2-step: resolve ID → query docs):

code
skill_mcp(mcp_name="context7", tool_name="resolve-library-id", arguments='{"libraryName": "react", "query": "hooks"}')
skill_mcp(mcp_name="context7", tool_name="query-docs", arguments='{"libraryId": "/facebook/react", "query": "useEffect"}')

Grep.app (search literal code patterns, not keywords):

code
skill_mcp(mcp_name="grep_app", tool_name="searchGitHub", arguments='{"query": "useActionState(", "language": ["TypeScript", "TSX"]}')

Tavily Search (real-time web search with topic filtering):

code
skill_mcp(mcp_name="tavily", tool_name="tavily_search", arguments='{"query": "Next.js 15 features", "max_results": 5}')
skill_mcp(mcp_name="tavily", tool_name="tavily_search", arguments='{"query": "OpenAI API pricing", "topic": "news", "days": 7}')
skill_mcp(mcp_name="tavily", tool_name="tavily_search", arguments='{"query": "AAPL stock", "topic": "finance"}')

Tavily Extract (get full content from specific URLs):

code
skill_mcp(mcp_name="tavily", tool_name="tavily_extract", arguments='{"urls": ["https://docs.example.com/guide"]}')
skill_mcp(mcp_name="tavily", tool_name="tavily_extract", arguments='{"urls": ["https://linkedin.com/in/someone"], "extract_depth": "advanced"}')

Tavily Crawl (crawl multiple pages from a starting URL):

code
skill_mcp(mcp_name="tavily", tool_name="tavily_crawl", arguments='{"url": "https://docs.example.com", "max_depth": 2, "limit": 10}')
skill_mcp(mcp_name="tavily", tool_name="tavily_crawl", arguments='{"url": "https://blog.example.com", "instructions": "Find all tutorial pages"}')

Tavily Map (discover site structure without extracting content):

code
skill_mcp(mcp_name="tavily", tool_name="tavily_map", arguments='{"url": "https://docs.example.com", "max_depth": 3}')

Exa:

code
skill_mcp(mcp_name="websearch", tool_name="web_search_exa", arguments='{"query": "Next.js 15 features", "numResults": 5}')

Common Mistakes

❌ Wrong✅ Correct
mcp_name="research-tools"mcp_name="context7"
tool_name="context7_resolve-library-id"tool_name="resolve-library-id"
tool_name="grep_app_searchGitHub"tool_name="searchGitHub"
tool_name="context7_get-library-docs"tool_name="query-docs"
tool_name="tavily_tavily_search"tool_name="tavily_search"