mcp-selection

name	mcp-selection
description	MCP tool selection guide — use when choosing between Kagi, Context7, gh_grep, Firecrawl, Playwright, or other MCP tools for a task. Provides decision matrix and anti-patterns.
keywords	["mcp","tool-selection","context7","kagi","gh_grep","firecrawl","playwright","documentation","web-search"]
license	MIT
metadata	{"priority":"high","replaces":"none"}

When to Load This Skill

Load this skill when you need to decide which MCP tool to use for a task. If you already know the right tool, skip this and call it directly.

MCP Server Management

Primary: Vision (vision)

• Use vision_vision_list to see all available MCP servers and their status
• Use vision_vision_add to add new MCP servers dynamically
• Use vision_vision_remove to remove servers
• Use vision_vision_status to check daemon health and uptime
• Use vision_vision_search to find servers in the catalog
• Use vision_vision_guidance to get tool selection recommendations

When to use: Managing MCP server lifecycle, checking what tools are available, troubleshooting server issues

Web Search & Research

Primary: Kagi (kagimcp)

• Use kagi_kagi_search_fetch for web searches, research, current information, news
• Use kagi_kagi_summarizer for summarizing web pages and documents
• Fast, high-quality results without tracking

Avoid: Using Playwright, Firecrawl, or general fetch tools for simple searches

Library & API Documentation

Primary: Context7 (context7)

• Use context7_resolve-library-id first to find the library ID
• Use context7_query-docs to get documentation for specific questions
• Best for: React, Next.js, TypeScript, Python libraries, etc.

Avoid: Web searching for documentation that Context7 has indexed

Code Examples & Patterns

Primary: Grep by Vercel (gh_grep)

• Use gh_grep_searchGitHub to find real-world usage examples on GitHub
• Great for: implementation patterns, API usage, seeing how others solved problems
• Filter by language with language (array), by repo with repo (string)

Avoid: Web searching for code examples when gh_grep can find them directly

Web Scraping & Data Extraction

Primary: Firecrawl (firecrawl) — typically available in standard deployments

• Use firecrawl_firecrawl_scrape for single page content extraction
• Use firecrawl_firecrawl_crawl for multi-page crawl (async — returns job ID)
• Use firecrawl_firecrawl_check_crawl_status to poll crawl job results

When to use over Kagi: When you need the full page content, structured data, or JS-rendered pages

Browser Automation (Playwright - if available)

Use ONLY for:

• E2E testing
• Filling out forms
• Clicking buttons
• Interactive web tasks
• Screenshots of rendered pages
• Exploring interactive application behavior

NEVER use for:

• Web search (use Kagi)
• Fetching page content (use Firecrawl or fetch)
• Research (use Kagi)
• Documentation lookup (use Context7)
• General web browsing

Playwright is for exploring application behavior, not browsing.

Decision Matrix

Task	Tool	Why
"Search for X"	Kagi	Purpose-built for search
"How do I use React hooks?"	Context7	Official documentation
"Show me examples of useEffect"	gh_grep	Real-world code
"Get content from example.com"	Firecrawl	Content extraction
"Summarize this article"	Kagi summarizer	Built-in summarization
"Find recent AI papers"	Kagi	General web search
"Click the login button"	Playwright	Browser automation

Anti-Patterns to Avoid

1. Don't use Playwright for research - It's slow and meant for automation
2. Don't web search for library docs - Context7 has them indexed
3. Don't scrape when you can search - Kagi is faster for finding info
4. Don't use fetch for complex pages - Firecrawl handles JavaScript rendering

Keywords

mcp tool selection, which tool to use, kagi vs context7, web search, documentation lookup, code examples, web scraping, browser automation, playwright, firecrawl, gh_grep, fetch, url content, tool choice, decision matrix