Brave Search is a Search MCP server that lets Claude Code, Cursor, Windsurf and any MCP-compatible AI agent web search via Brave Search API. Install in 1 minute with mcpizy install brave-search.
Search
Web search via Brave Search API
mcpizy install brave-searchnpx -y @modelcontextprotocol/server-brave-searchBrave Search MCP exposes Brave's independent web search index as agent tools: `web_search`, `local_search`, `news_search`, `image_search`. Authentication is a Brave Search API key (free tier: 2,000 queries/month; paid tiers from $3 per 1k extra queries). The MCP returns results with title, URL, description, and an `age` field that helps agents reason about content freshness.
We use Brave Search MCP for three workflows. First, agent-driven research: "find the three most recent posts about Claude Agent SDK pricing changes" — `web_search` with a date filter returns a ranked list, the agent reads the top hits via Fetch MCP. Second, local-business lookups via `local_search` (powered by Brave's POI index — surprisingly good outside US). Third, news-sensitive workflows: "what's the latest on the EU AI Act enforcement timeline" — `news_search` returns articles from the last week. Token cost is lean — 10 results with descriptions is ~2-3k tokens.
Compared to Google Search (no official MCP) or Bing Search MCP, Brave wins on price (much cheaper per query), privacy (no user tracking), and an independent index (not a Google-API reseller). The trade-off is index size: Brave's index is smaller than Google's; rare long-tail queries return fewer hits. For everyday research and news, the gap is invisible. Compared to Perplexity MCP, Brave returns raw search results (the agent reads the underlying pages); Perplexity returns a synthesised answer. Different layer of abstraction.
Free-tier quota is 2,000 queries/month — easy to exhaust if the agent searches aggressively. Pin the agent to "search once, read top 3 results" patterns rather than "search every variation of the query."
`web_search` results' `age` field is approximate, not exact. For freshness-critical queries, pair Brave Search with Fetch MCP and read the article's actual publish date.
Brave's index has known gaps in some non-English languages — French, German, Spanish are well-covered; Japanese, Arabic, Hindi have thinner coverage. For non-English research, fall back to Bing Search or Google via Serper.
Result descriptions are snippets, not full pages. Agents that try to answer based on description alone often hallucinate; the answer is to read the underlying URL via Fetch MCP.
Honest pros/cons against the closest search MCP servers.
| Server | Strengths | Trade-offs |
|---|---|---|
| Perplexity MCP | Returns synthesised answers with citations, less agent work | Pricier, less control over which sources are used |
| Bing Search MCP (community) | Larger index than Brave for long-tail queries | Pricier per query, Microsoft tracking |
| Serper MCP (community) | Google results via SerpAPI-style proxy | Most expensive option, terms-of-service grey area |
Privacy-first web search engine
Full-text and semantic search engine. Lightning fast, typo-tolerant search with filtering and faceting.
Search the web using Kagi's premium, ad-free search API. High-quality results without tracking.
RAG search over your documentation and content. AI-powered knowledge base.
If Brave Search doesn't fit your stack, these Search MCP servers solve similar problems.
The Brave Search MCP server is an Search Model Context Protocol server that lets Claude Code, Cursor, Windsurf, VS Code with Copilot, and other MCP-compatible AI agents web search via Brave Search API. It exposes Brave Search's capabilities as tools the AI can call directly from your editor or CLI.
The fastest way is the MCPizy CLI: run `mcpizy install brave-search` and MCPizy will add the server to your `.claude.json` automatically. You can also install it manually by adding an entry under `mcpServers` in `.claude.json` with the command `npx -y @modelcontextprotocol/server-brave-search` and restarting Claude Code.
Yes. The Brave Search MCP server is free and open source (see the GitHub repository linked on this page). You may still need a Brave Search account or API key to connect the server to the underlying service, but the MCP layer itself has no MCPizy subscription cost.
Yes. Any MCP-compatible client works — including Claude Code, Claude Desktop, Cursor (via `.cursor/mcp.json`), Windsurf, VS Code with Copilot Chat, and custom agents built on the MCP SDK. The same install command targets all of them; only the config file path differs.
Once installed, your AI agent can web search via Brave Search API directly inside your conversation. Typical use cases include asking Claude Code or Cursor to run Brave Search operations, inspect results, chain Brave Search with other MCP servers (see our Workflow Recipes), and automate repetitive search tasks without leaving your editor.