Brave Search MCP Server

The Brave Search MCP server is the official integration from Brave for connecting AI agents to the Brave Search API. It exposes Brave's independent search index (not a Google or Bing wrapper) through eight tools spanning web search, local business search, points of interest, image and video search, news, AI-powered summarization, and a dedicated LLM context tool that returns pre-extracted web content for RAG pipelines.

The server is written in Node.js (requires Node 22.x or higher) and supports both STDIO (default) and HTTP transports, making it usable with Claude Desktop, Cursor, VS Code, and remote MCP clients. It is distributed as the @brave/brave-search-mcp-server npm package and also publishes a Docker image. Tool exposure can be filtered with BRAVE_MCP_ENABLED_TOOLS and BRAVE_MCP_DISABLED_TOOLS environment variables, and logging is configurable from debug through emergency levels.

Because the server is maintained by Brave itself, tool coverage tracks the Brave Search API closely. A paid Brave Search API plan is required for some capabilities (Pro features unlock local search and extended snippets), and all requests authenticate via a BRAVE_API_KEY environment variable obtained from the Brave developer dashboard.

Tools

Prerequisites

• Node.js 22.x or higher
• A Brave Search API key from the Brave Search API dashboard (free and paid tiers available; local search requires a Pro plan)

Install via NPX (recommended)

Add the server to your MCP client config. For Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@brave/brave-search-mcp-server"],
      "env": {
        "BRAVE_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

Run via Docker

docker build -t mcp/brave-search:latest .

{
  "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"
      }
    }
  }
}

Build from source

git clone https://github.com/brave/brave-search-mcp-server.git
cd brave-search-mcp-server
npm install
npm run build
node dist/index.js                          # STDIO (default)
node dist/index.js --transport http --port 8000   # HTTP transport

Optional environment variables

• BRAVE_MCP_ENABLED_TOOLS: comma-separated allowlist of tool names
• BRAVE_MCP_DISABLED_TOOLS: comma-separated denylist of tool names
• Logging level can be set from debug through emergency

• Real-time web research and fact-checking for an AI agent using an index independent of Google or Bing
• Pulling fresh news articles into briefings or daily digests with the news freshness filter
• Powering local discovery experiences (restaurants, shops, services) via local and place search
• Building RAG pipelines that need pre-extracted web content via brave_llm_context without writing a custom scraper
• Generating cited AI summaries of search results with brave_summarizer for research workflows

• "Search the web for the latest benchmarks on Llama 3.3 70B and summarize the top three sources."
• "Find news articles from the past day about EU AI Act enforcement."
• "Get me the top-rated coffee shops within 1km of latitude 37.7749, longitude -122.4194."
• "Pull image search results for 'aurora borealis Iceland March 2026' and list the top 10."
• "Use Brave LLM context to fetch up to 20 URLs about vector database benchmarks and return clean text for my RAG index."

• Officially maintained by Brave, so tool coverage tracks the Brave Search API directly
• Independent search index, useful when you want results not derived from Google or Bing
• Broad surface area: web, news, images, video, local, places, summarization, and LLM-ready context in one server
• Supports both STDIO and HTTP transports, plus npm and Docker distribution

• Requires a Brave Search API key, and some tools (local search, extended snippets) need a paid Pro plan
• Requires Node.js 22.x or higher, which can be a friction point on older systems
• Summarization is two-step: you must run a web search first to obtain the key before calling brave_summarizer

• Tavily MCP server: search and extract API focused on LLM/RAG use cases
• Exa MCP server: neural search API with content extraction
• Perplexity Ask MCP server: answers backed by Perplexity's search and summarization stack