Brave Search MCP

Overview

Brave Search MCP wraps the Brave Search API as a Model Context Protocol server. Drop in, set a key, and any MCP-compatible agent gets clean web results with no Google API quota games. The reference implementation lives in the MCP servers monorepo and is the most-installed search server.

Best for

Best for
Any agent that needs general web search without operating its own crawler

Use cases

• Give a Claude or Cursor agent real web search without a custom backend
• Bootstrap a research agent that needs general web context
• Replace 'agent invents URLs' with 'agent searches first'
• Mix with a fetch tool for full search-and-read flows

How to use

Install

npx -y @modelcontextprotocol/server-memory

Tested with

Claude Desktop

Example client config

{\n  "mcpServers": {\n    "brave-search": {\n      "command": "npx",\n      "args": ["-y", "@modelcontextprotocol/server-brave-search"]\n    }\n  }\n}

Notes

Why it matters

Every agent eventually needs to search the web. Brave Search MCP is the default way to plug that in without standing up a custom search service.

How teams use it in production

Pair with the webpage info skill (or a browser MCP for blocked sites) for a full search-and-read loop. Cache results aggressively to keep costs flat.

What to watch

The space of search MCPs (Tavily, Serper, Exa, Brave) is consolidating quickly. Brave wins on privacy posture and price floor today.