MCP Spotlight: Brave Search — Give Your Agent Real-Time Web, News, Video & Local Search

Server: @brave/brave-search-mcp-server by Brave Stars: 1.6K+ · Downloads: 7,160+ · Tools: 6 · License: Open source MCP Tracker: glama.ai/mcp/servers/brave/brave-search GitHub: brave/brave-search-mcp-server

Grounding an LLM in real-time information was one of the first problems MCP was designed to solve. Brave Search MCP is the cleanest answer: 6 tools covering web, news, image, video, and local search, plus AI-powered summarization, all through a single API key and a single npx command.

The Six Tools

Tool	What It Does	Key Differentiator
brave_web_search	Full web search with rich result types and advanced filtering	Country, language, freshness ("last week", "last month"), safesearch, spellcheck, pagination, custom re-ranking via Goggles
brave_local_search	Local business and place search with ratings, hours, AI descriptions, review sentiment	Pro plan unlocks full local search; falls back to web search on Free tier
brave_image_search	Image search returning URLs and metadata	v2 removed base64 encoding — 10× faster responses, no context-window bloat
brave_video_search	Video search with comprehensive metadata and thumbnails	Up to 50 results per query with freshness filtering
brave_news_search	News articles with freshness controls and breaking-news indicators	Defaults to last 24 hours; Pro unlocks extra snippets
brave_summarizer	AI-generated summaries from web search results	Two-step flow: search with summary: true, then call summarizer with the returned key

Search That Knows What Your Agent Needs

Brave Search MCP isn't a thin wrapper around a REST API. The tools are designed for agent workflows:

Rich filtering built into every tool:

• country — scope results to a specific country (US, DE, FR, etc.)
• search_lang — filter by content language, not just UI language
• freshness — "past day", "past week", "past month", "past year", or exact date ranges
• safesearch — "off", "moderate", or "strict"
• spellcheck — automatically corrects typos in queries
• goggles — custom re-ranking definitions for domain-specific result weighting

This means your agent doesn't post-process raw results. It asks for exactly what it needs and gets structured output it can reason about directly.

Two-Step Summarization

The brave_summarizer tool uses a deliberate two-step pattern:

1. Search: call brave_web_search with summary: true → get summary keys alongside results
2. Summarize: call brave_summarizer with the key → get an AI-generated summary of the search results, optionally with inline source references and entity information

This separation matters: the agent sees the raw results, decides which summary key is relevant, and then calls the summarizer. The summarizer doesn't replace the search — it augments it. The agent retains full visibility into what it's summarizing.

v2 Migration: What Changed

Version 2 of the MCP server made two important changes:

1. STDIO is now the default transport — following MCP conventions. HTTP is still available via --transport http or BRAVE_MCP_TRANSPORT=http.

2. Image search no longer returns base64 — v1 encoded every image as base64 in the response, which bloated responses and ate context windows. v2 returns image URLs and metadata only, matching the native Brave Search API response shape. Your agent can then fetch images as needed.

These changes reflect production learnings: context windows are precious, and STDIO is the established MCP default.

Authentication: One API Key

# Get your key at https://api-dashboard.search.brave.com/app/keys
# Free: 2,000 queries/month
# Pro: extra snippets, local search, AI summaries

Facio Integration

STDIO transport (default, recommended):

{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@brave/brave-search-mcp-server"],
      "env": {
        "BRAVE_API_KEY": "${credentials.BRAVE_API_KEY}"
      }
    }
  }
}

Docker (for containerized deployments):

{
  "mcpServers": {
    "brave-search": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "BRAVE_API_KEY", "docker.io/mcp/brave-search"],
      "env": {
        "BRAVE_API_KEY": "${credentials.BRAVE_API_KEY}"
      }
    }
  }
}

Facio's audit trail captures every search query alongside the agent's reasoning. This creates traceability from "the agent recommended this approach" back to "because it searched for X and found Y." For compliance-sensitive environments, this closes the loop.

Quickstart

# One-click install via Smithery
npx -y @smithery/cli install brave

# Or manual with Claude Desktop / VS Code:
# Add to your MCP config (see above)

# First prompts after connection:
# "Search the web for the latest MCP protocol specification changes"
# "Find news about AI regulation in the EU from the past week"
# "Search for local coffee shops near Berlin with ratings above 4.5"
# "Search for 'gradient descent visualization' images — just URLs, no base64"
# "Search the web for 'WebAssembly performance benchmarks', then summarize the results"

Default language is en-US, but search_lang and country parameters let you scope queries globally.

When Brave Search MCP Makes Sense

Brave Search MCP is the right choice when your agent needs:

• Real-time web information without hallucination risk — the agent cites sources, it doesn't guess
• Structured search with agent-friendly filtering — country, language, freshness, safesearch — all as tool parameters, not post-processing
• Multi-modal search from one server — web, news, images, video, local, and AI summaries in a single MCP endpoint
• Privacy-respecting search — Brave's independent index, not dependent on Google or Bing

For agents that need to answer questions about the real world, verify facts, research current events, or find local businesses, this is the foundation layer.

Bottom Line

Brave Search MCP gives AI agents what they've been missing: real-time, structured access to the web, news, images, video, and local search through a single, well-maintained MCP server. Six tools, 2,000 free queries per month, STDIO transport, and official Brave maintenance.

At npx -y @brave/brave-search-mcp-server, it's the fastest way to ground any MCP-compatible agent in real information.

---

MCP Spotlight is a series covering servers that give AI agents real capabilities. Every server is evaluated for tool quality, integration depth, and fit with Facio's HITL-first agent runtime.