Brave Search MCP Server (Web Search via API)
Set up @modelcontextprotocol/server-brave-search with a BRAVE_API_KEY so Claude or Cursor can perform live web lookups.
Detailed Explanation
Brave Search Server: Live Web Lookups
Claude Desktop ships without web access by default. The Brave Search MCP server adds it via Brave's independent search index — a privacy-respecting alternative to Google or Bing.
Config
{
"mcpServers": {
"brave-search": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-brave-search"],
"env": {
"BRAVE_API_KEY": "BSA_xxxxxxxxxxxxxxxxxxxx"
}
}
}
}
Getting an API key
- Visit api.search.brave.com and create an account.
- Subscribe to the Free tier (2,000 queries/month, no credit card required) or a paid plan if you need more.
- Copy the API key from the dashboard. It starts with
BSA_. - Paste it into the env field. The key authenticates against Brave's API, not your local browser session.
Tools the server exposes
The server registers two MCP tools: brave_web_search for general queries and brave_local_search for location-scoped lookups (restaurants, businesses, etc.). The model decides which to call based on the user's question.
Quota tracking
The free tier resets monthly; if you exceed it the server returns a 429 error and the model will see a tool call failure. Monitor usage on the Brave API dashboard, or check our rate-limit calculator to estimate how many queries 2,000/month means in practice.
Why not just use a fetch server?
You could approximate web search with the fetch server pointing at Brave or DuckDuckGo's HTML endpoint, but parsing the HTML response is brittle. The dedicated MCP server returns clean JSON results and handles pagination and rate-limit headers for you.
Use Case
Asking Claude Desktop 'what's the latest stable Node.js LTS version?' or 'find blog posts about MCP servers from this month' and getting answers grounded in current web results rather than a 2024 training cutoff.