Brave Search is the privacy-first web index — independent of Google, no user tracking, results ranked without ad-bid distortion. The skill exposes Brave’s API as an agent tool: web search, news search, image search, code search verticals, all accessible without the SEO spam that clogs Google’s top 10.
What it produces: web_search, news_search, images_search, code_search — each returning ranked results with titles, URLs, snippets, freshness signals. Locale + safesearch params per call.
Best for: general-purpose web search inside an agent loop where you want clean, recent, lower-spam results. Especially good for technical lookups (where Brave’s code-search vertical surfaces GitHub, StackOverflow, docs) and news (where the freshness ranking actually works).
Skip if: you need semantic search (“articles like this one”) — use Exa Search for that. Skip if you need Google-specific result patterns (PAA boxes, AI Overviews) — Brave is its own index, not a Google clone.
Setup gotchas: API key required. Free tier is 2,000 queries/month; “Pro” starts at $5/month for 20K. Enough for most agent workloads; a busy research agent can blow through 2K in days. Set BRAVE_API_KEY. The independent-index thing is real — don’t expect identical results to Google for the same query.
Real-world workflow: my “default search” inside Claude Code. Whenever the agent needs to look something up that isn’t in Context7 (which covers libs) and isn’t a known URL (which uses Fetch MCP), it goes to Brave. Cleaner results, lower noise, fewer follow-up reads.
Compatible alternatives: Exa Search for semantic neural search, Firecrawl MCP for site-deep crawls.
The cleanest “general search” layer in the MCP ecosystem. Worth the $5.