Chaining tools beats a single call. Each recipe below is a concrete sequence; pick the variant that fits your goal.

search_web(query="Latest NVIDIA H200 benchmarks")
  → fetch_page(url="<top result>")
  → retry with search_web(query="...", provider="exa") for JS-heavy pages

• Start broad with search_web (auto provider tries DDG, falls back to Exa).
• Use fetch_page to pull full content from the most relevant hits.
• Exa's semantic ranking surfaces pages DuckDuckGo misses; its JS fallback handles SPAs.

search_github(query="uv package manager")
  → get_github_issue(url="<top issue/PR URL>")

• search_github finds relevant issues/PRs across repos.
• get_github_issue then pulls the full thread, sorted by reactions — highest-signal discussion first.

search_web(query="<exact error string>")
  → fetch_page(url="<relevant Stack Overflow / docs result>")

• Search the literal error first (wrapped in quotes works best).
• fetch_page the most relevant result to read the fix in context.

search_reddit(query="...")
  + search_hackernews(query="...")
  + search_x(query="...")

• Triangulate sentiment across community (Reddit), developer (HN), and real-time (X) sources.
• Note: search_x requires auth; the other two are keyless.

search_arxiv(query="au:Goodfellow AND cat:cs.LG")
  → search_wikipedia(query="<concept>") for background

• Use Lucene prefixes (au:, ti:, cat:, abs:) to target arXiv fields.
• Fill in encyclopedic background with search_wikipedia.

search_linkedin(query="site reliability engineer", content_type="people")
  → search_linkedin(query="machine learning startup", content_type="companies")

• content_type filters to people / companies / posts / articles / jobs.
• Keyless via DuckDuckGo site:linkedin.com + Jina Reader enrichment.