Brave Search

Brave maintains this server in-house at brave/brave-search-mcp-server. It replaces the archived @modelcontextprotocol/server-brave-search package and is the supported way to give an agent access to Brave's search index. Useful as a second-opinion source in research agents because the index is independent of Google and Bing. What it does: - Web search over Brave's own crawled index, with site, freshness, and country filters - News search returning headline, publisher, and publication time - Image and video search with safe-search and resolution filters - Local POI search for business names, hours, and geographic coordinates - Optional AI summarisation via the Brave Summarizer API for long-form answers - Honest provenance — every tool response includes the originating URL for citation Example: Give a research agent access to BRAVE_API_KEY and it can combine "web search Brave" and "news search Brave" in the same session, cross-checking a trending claim against independently-crawled sources before it writes the final answer. Setup: BRAVE_API_KEY env var is the only required config. The free tier allows a few thousand queries per month; paid tiers lift the rate cap and unlock the Summarizer endpoint. Configure BRAVE_SEARCH_COUNTRY if you need region-locked results. Transport: stdio for local use, plus a Streamable HTTP variant for hosted deployments. Notes: No upstream scraping or Google fallback — answers reflect Brave's own crawler, which trades recall against Google for independence and privacy. Rate-limit headers are surfaced as MCP tool errors so a client can back off gracefully rather than silently burning quota.