Brave Search MCP Server: Web Search for Claude

The Brave Search MCP Server is Brave’s official MCP integration for the Brave Search API. It gives Claude an independent web index with several specialized tools: brave_web_search, brave_local_search, brave_place_search, brave_video_search, brave_image_search, brave_news_search, brave_summarizer and brave_llm_context, which returns content shaped for language models.

Who it is for

Anyone who wants Claude Code or Claude Desktop to search the live web with a privacy-focused provider, and builders who need filters like freshness, country, language and safe search.

Install

Get an API key from the Brave Search API dashboard. The README’s Docker config for Claude Desktop:

{
  "mcpServers": {
    "brave-search": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "BRAVE_API_KEY", "docker.io/mcp/brave-search"],
      "env": { "BRAVE_API_KEY": "YOUR_API_KEY_HERE" }
    }
  }
}

With npx in Claude Code (stdio is the default transport):

claude mcp add brave-search -e BRAVE_API_KEY=YOUR_API_KEY -- npx -y @brave/brave-search-mcp-server

Use BRAVE_API_KEY_FILE to read the key from a mounted secret instead of an environment variable.

Use cases

• Research a library’s latest release notes or a breaking change.
• Pull recent news with the freshness filter for a weekly briefing.
• Find local businesses or places with ratings and hours.
• Feed brave_llm_context results into a summarization prompt.

Pros

• Official, MIT licensed and actively maintained.
• Independent index rather than a reseller of another engine.
• Tool whitelists keep the context footprint small.

Cons and caveats

• Requires a paid or quota-limited API plan.
• Some parameters, such as extra snippets, are Pro-plan only.
• The HTTP transport binds to loopback by default; review host and origin settings before exposing it.

Links

• Repo: brave/brave-search-mcp-server
• Alternatives: Exa MCP, Firecrawl MCP
• Guide: best MCP servers