The Brave Search MCP server connects Claude to Brave’s web search API. The version Anthropic once published is archived and its npm package is deprecated. Brave now maintains its own server, @brave/brave-search-mcp-server. It gives Claude live access to the web for queries that need fresh information: current events, recent product releases, anything that changed after the model’s knowledge cutoff.
This is a low-friction search server to start with. Brave’s monthly credit covers light personal use, and the API key comes from Brave’s developer dashboard.
Why use it
Models go stale. Adding a search MCP server gives Claude a way to ground its answers in current information. Without one, Claude either hallucinates recent events or refuses to answer them.
Brave specifically wins on privacy. Their index is independent and doesn’t profile users. The monthly credit is enough for light use. For teams or heavy usage, check Brave’s paid pricing against alternatives.
What it actually does
Brave’s own server offers several search tools, including web and local search, with more for images, video and news. Web search returns the standard list of titles, URLs, and snippets that Claude reads and synthesises. Local search returns places near a location.
Practical patterns:
- “What’s the latest news on the EU AI Act?”
- “Find a recent article comparing Claude 4 vs GPT-5 benchmarks.”
- “Search for cafés near King’s Cross station.”
Gotchas
The monthly credit is small. Heavy daily research can use it up quickly. If you’re using Claude as a daily research assistant, plan for paid usage or pair Brave with Fetch so Claude can read full articles after a single search.
Search snippets are limited to a couple of lines. For deep reading, Claude needs to fetch the URL separately. Pair Brave Search with the Fetch server for end-to-end “search then read” workflows.
Quality varies by query type. Brave is strong on tech and news, weaker on shopping or hyperlocal queries than Google. Combine with Tavily or Perplexity for a richer search stack.
