deep-research

Break the topic into 3-5 sub-questions, search each one with firecrawl or exa MCP tools, deep-read 3-5 key sources, then synthesize a structured report with inline citations and a source list. Every claim should link to a source.

At least one of firecrawl or exa MCP servers is required. Firecrawl provides firecrawl_search, firecrawl_scrape, and firecrawl_crawl; exa provides web_search_exa, web_search_advanced_exa, and crawling_exa. Using both together gives the best coverage.

Firecrawl focuses on searching, scraping, and crawling pages into clean content, while exa offers semantic web search with filters like publication date and token-limited crawling. The skill supports either, and combining both improves source coverage.

Yes. For broad topics, launch parallel research agents with the Task tool, assigning each agent a subset of sub-questions. Each agent searches and reads sources independently, and the main session synthesizes their findings into the final report.

Reports depend on available search results and source quality, so niche topics may have insufficient data. Single-source claims are flagged as unverified, and gaps in coverage are explicitly acknowledged rather than filled with speculation.