Brave Search MCP Server

Generates AI-powered summaries from web search results using Brave's summarization API.

Workflow:

1. Perform a web search using brave_web_search and set the summary parameter to true.
2. Extract the returned summary key from the results.
3. Pass that key to the brave_summarizer tool.

Parameters:

• key (string, required): Summary key from web search results
• entity_info (boolean, optional): Include entity information (default: false)
• inline_references (boolean, optional): Add source URL references (default: false)

Docker Build

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

Local Build

npm install
npm run build

To use the Brave Search MCP Server, you must obtain an API key from Brave:

1. Sign up for a Brave Search API account.
2. Choose a plan:
   • Search: Provides real-time search data (URLs, text, news, images, etc.) optimized for AI.
   • Answers: Provides summarized, grounded answers to questions.
3. Generate your key from the developer dashboard.

For local development with Docker, use Docker Compose:

docker-compose up --build

Ensure BRAVE_API_KEY (or BRAVE_API_KEY_FILE) is set in your shell or a .env file.

Using Docker Compose Secrets

To avoid plain-text environment variables, use Docker Compose secrets:

1. Prepare the secret file:

cp secrets/brave_api_key.txt.example secrets/brave_api_key.txt

2. Add your key to secrets/brave_api_key.txt.
3. Start the stack with the secrets override:

docker compose -f docker-compose.yml -f docker-compose.secrets.example.yml up --build

By default, the Brave Search MCP server uses STDIO transport. If you need to use HTTP transport, you must explicitly configure it using one of the following methods:

1. Set the BRAVE_MCP_TRANSPORT environment variable to http.
2. Provide the --transport http runtime argument when launching the server.

To use the server with Claude Desktop, add a configuration entry to your claude_desktop_config.json.

Using Docker

{
  "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"
      }
    }
  }
}

Using NPX

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

To test the server using the MCP Inspector:

1. Build and start the server:

npm run build
node dist/index.js

2. In a separate terminal, start the Inspector:

npx @modelcontextprotocol/inspector node dist/index.js

Note: For HTTP mode testing, add --transport http to the arguments in the Inspector UI.

The Query interface provides context about the search request. It includes the original query string and an altered version if the engine modified it. It also provides metadata such as:

• safesearch: Boolean indicating if safesearch was active.
• is_geolocal: Whether the query was treated as location-sensitive.
• language: The detected language of the query.
• country: The country used for the request.
• is_trending: Whether the query is currently trending.
• lat/long/city/state: Geographic context gathered from the query.

Each image result in a brave_image_search response provides several layers of metadata:

• thumbnail: Contains a src URL and dimensions (width, height) for a smaller preview.
• properties: Contains the direct url of the image, a placeholder (lower resolution version), and the image's dimensions.
• meta_url: Provides structural information about the host URL, including the scheme, hostname, path, and the favicon URL.
• confidence: An enum indicating the search engine's confidence in the result: low, medium, or high.