# mikechao/brave-search-mcp [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/mikechao/brave-search-mcp  
**GitHub Stars:** 130  
**npm Downloads (last month):** 1251  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/mikechao-brave-search-mcp

## Description
Web, Image, News, Video, and Local Point of Interest search capabilities using Brave's Search API

## Tools
Capabilities this server exposes over MCP:

- **query** — (string): The term to search for videos
- **count** — (number, optional): The number of videos to return (max 20, default 10)

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "brave-search-mcp": {
    "command": "npx",
    "args": ["-y","brave-search-mcp"],
    "env": {
      "BRAVE_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `BRAVE_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What mikechao/brave-search-mcp MCP server does

The mikechao/brave-search-mcp MCP server exposes Brave Search API operations as MCP tools. It supports ordinary web searches, image discovery, news lookup, video searches, and searches for local businesses, services, restaurants, and other points of interest. Local search can fall back to a web search when the initial request does not produce location results.

Its context-oriented search tool is intended for agent grounding and retrieval-augmented generation. It can return extracted snippets from matching pages, optionally target one exact URL, and limit the response by result count, URL count, snippets, tokens, and output size. Compact mode applies relevance filtering and local truncation; full mode returns raw snippets without those local reductions.

## How it works

The process receives MCP tool calls and sends corresponding requests to Brave Search API. The default transport is stdio, which suits desktop and editor clients that launch local MCP processes. An HTTP flag enables Streamable HTTP mode instead, with the server listening on port 3001 by default at `/mcp` and binding to `0.0.0.0` unless configured otherwise.

Search tools support result limits, and several support pagination through an offset. Web, news, and video searches can filter results by discovery date, including predefined periods or a custom date range. The image, local, and video tools have their own result limits.

## Setup and configuration

The mikechao/brave-search-mcp MCP server requires a Brave Search API account, a selected plan, and an API key from the Brave developer dashboard. Set `BRAVE_API_KEY` before starting it:

```bash
BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp
```

Use HTTP mode with:

```bash
BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp --http
```

`PORT` changes the HTTP port, while `HOST` changes the bind interface. HTTP deployments can restrict accepted hostnames with `ALLOWED_HOSTS`. Feature settings may be supplied through environment variables or through a TOML file selected by `BRAVE_MCP_CONFIG`; file mode makes that file the source of truth for overlapping feature settings.

Optional controls include request limits, cooldown windows, policy files, audit logging, query redaction, and requiring a justification string on tool calls. These settings are available in environment mode as documented by the project.

## Tools and capabilities

- `brave_web_search`: web results with count, pagination, and freshness filters.
- `brave_image_search`: image results, with up to 50 results requested.
- `brave_news_search`: news results with pagination and freshness filters.
- `brave_video_search`: video results with pagination and freshness filters.
- `brave_local_search`: local businesses, services, and points of interest.
- `brave_llm_context_search`: extracted web snippets for AI grounding, with URL, token, snippet, and response-mode controls.

The project also documents OpenAI Apps and MCP Apps support. When UI mode is enabled for a tool, a corresponding widget can control what is added to model context.

## Limitations and notes

All searches depend on Brave Search API access and the limits of the selected API plan. Individual tools cap result counts and pagination values, while context search enforces bounds on queries, tokens, URLs, snippets, and output size. The context query is limited to 400 characters and 50 words. HTTP mode should be configured deliberately, especially its bind host and allowed-host list, when exposed beyond a local process.

_Full upstream README: https://allmcps.com/mcp/mikechao-brave-search-mcp/readme_