MCP ServerSTDIO

Brave Search MCP Server

The Brave Search MCP server, from the official Model Context Protocol servers project, gives an AI model live web search through a single Brave Search API key. It is a privacy-focused alternative to Google search that returns fresh results without tracking — point a model at it and it can research the open web in real time.

Open Agent Studio modelcontextprotocol/servers — brave-search

Hosted URL

Bring your own

Suggested model

Claude Sonnet 4.5

MCP Playground runs 40+ models on the same workflow: switch anytime, or use Compare mode to run several in parallel and balance quality vs. cost.

Auth

A single Brave Search API key (BRAVE_API_KEY). Grab one free at brave.com/search/api, or deploy a sandboxed hosted instance from the MCP Playground catalog and paste the key at deploy time.

What the Brave Search MCP server does

How models use it and what it is built for.

The Brave Search MCP server wraps the Brave Search API as two MCP tools. Models can run a general web search with pagination and freshness filters, or a local search for businesses and places with addresses, ratings and hours (falling back to web results when no local match is found). It runs as a Node.js stdio process — `npx @modelcontextprotocol/server-brave-search` — configured with a single BRAVE_API_KEY. Brave offers a free API tier, so it is a low-friction way to give any agent current, citeable web results instead of relying on the model's training cutoff.

Tools the Brave Search MCP server exposes

Typical tools an AI model can call. Exact names vary by version.

brave_web_search — run a general web search with pagination and freshness controls
brave_local_search — search for local businesses and places, with web fallback

Example prompts to try

Copy any of these into MCP Agent Studio after connecting.

Search the web for the latest Model Context Protocol spec changes in 2026 and summarise what's new.
Find three recent articles comparing vector databases and list their key claims with links.
Do a local search for well-reviewed coffee shops near King's Cross, London and rank them by rating.
What are people saying this week about the new GPT-5.6 models? Cite your sources.

Models on MCP Playground

This is not a single-model product: you get the same MCP connection with 40+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.

Default pick for Brave Search

Claude Sonnet 4.5

Claude Sonnet 4.5 chains searches, follows links and synthesises cited answers well. Drop to Haiku 4.5 for single-shot lookups to keep costs low.

Open Compare mode (side-by-side)Try this server in Agent Studio

Try the Brave Search MCP server in your browser

Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 40+ models, and start chatting — no install required.

Open Agent Studio

Brave Search MCP server — FAQ

Common questions about connecting, scoping and using it safely.

What is the Brave Search MCP server?

It is an open-source MCP server from the official Model Context Protocol servers project that lets AI models search the web through the Brave Search API. You give it a Brave API key and it exposes brave_web_search and brave_local_search tools to any MCP-compatible client.

Which tools does the Brave Search MCP server expose?

Two tools: brave_web_search for general web queries with pagination and freshness filters, and brave_local_search for businesses and places (addresses, ratings, hours), which falls back to web results when there is no local match.

Do I need a paid Brave Search API plan?

No. Brave offers a free Search API tier that is enough to evaluate and run light workloads. You only need a paid plan for higher query volumes; the MCP server works the same either way.

How is Brave Search different from a Perplexity or Tavily MCP server?

Brave returns raw ranked search results (links, snippets, local listings) and lets the model do the reading and synthesis. Perplexity returns a finished, cited answer, and Tavily is tuned for extraction and crawling. Many agents use Brave for breadth and Perplexity or Tavily for depth.

How do I test a Brave Search MCP server in the browser?

Open MCP Agent Studio, deploy the hosted Brave Search server from the catalog (or connect your own), paste your Brave API key, pick a model and start chatting. You can watch each brave_web_search call and its results live without any local setup.