Skip to content

Brave Search MCP Server

Brave's official MCP server for its own search index: web, news, images, video, places, summaries and LLM-ready extracted page text.

Reviewed by Sheela · Updated

Setup docs Pay per use

Our rating

4.3/5

A sound default search tool for agents, chiefly because brave_llm_context returns readable page text on the first call instead of ten links to fetch.

What it is

Brave's server wraps the Brave Search API. It runs locally with npx -y @brave/brave-search-mcp-server (version 2.1.4, released 17 September 2026, with about 94,000 npm downloads in the month to 18 October) or the mcp/brave-search Docker image, over stdio by default or HTTP with --transport http. You supply a BRAVE_API_KEY, or a key file for Docker secrets.

There are eight tools: brave_web_search, brave_news_search, brave_image_search, brave_video_search, brave_local_search, brave_place_search, brave_summarizer and brave_llm_context. The last returns pre-extracted page content sized for a model's context, with a budget between 1,024 and 32,768 tokens.

Tools can be switched on or off with BRAVE_MCP_ENABLED_TOOLS and BRAVE_MCP_DISABLED_TOOLS. In HTTP mode the server binds to 127.0.0.1 by default and rejects foreign browser origins, but the endpoint has no authentication of its own, so the README warns against exposing it outside a trusted network.

Install

npx -y @brave/brave-search-mcp-server --transport stdio

Our review

Brave is one of the few search APIs with its own web index rather than resold Google or Bing results, and its MCP server exposes nearly all of it. That matters more from next year: Brave's API page points out that Google's Custom Search JSON API shuts down on 1 January 2027, which leaves fewer independent options for agents that need general web search.

brave_llm_context is the tool that justifies the choice. It returns extracted text from the top results within a token budget you set, from 1,024 to 32,768, with caps per URL, so the model does not have to fetch and parse pages itself. Brave's own FreshQA run claims 85.3% accuracy at $18 per 1,000 questions including model tokens, against $49 to $89 for Parallel, Exa and Tavily. It is a vendor benchmark and should be read as one.

Pricing is simple. The Search plan costs $5 per 1,000 requests with $5 of credit each month, so the first 1,000 searches are free; past that, an agent making three searches per task across 10,000 tasks a month spends about $145. Answers, which powers summaries, adds token fees and allows only 2 requests a second against Search's 50. A card is required even for the free credit, which Brave says prevents abuse. Brave states the API is SOC 2 Type II attested and that queries are never shared or trained on; contractual zero data retention is an Enterprise term.

The server itself is local only. Brave hosts no endpoint, so a team wanting one shared instance runs it over HTTP, and that endpoint has no authentication: the 127.0.0.1 default is safe, while the 0.0.0.0 setting needed inside containers is not. Every result is untrusted web text arriving in the model's context, the usual prompt-injection exposure of any search tool, made sharper here because llm_context hands over page bodies rather than snippets.

Where it earns its place

  • A solo developer building a research agent on a small budget

    Ground answers in fresh web pages without writing a scraper

    Call brave_llm_context with a 4,096-token budget instead of search-then-fetch, which saves a round of page fetching per question. At $5 per 1,000 requests, 20,000 queries a month costs about $95 after the $5 monthly credit.

  • An engineering team whose agents use Google's Custom Search JSON API

    Replace the search backend before Google shuts it down on 1 January 2027

    Swap the agent's search tool for brave_web_search; Brave publishes a migration guide. The Search plan's 50 requests a second covers most internal agents, and higher capacity or custom terms go through Brave's Enterprise sales.

  • A local-listings startup

    Find businesses and opening hours around a map coordinate

    brave_place_search takes latitude, longitude and a radius in metres and returns addresses, hours, ratings and photos for up to 50 places per call, drawn from the same prepaid credits as web search.

Before you commit

Setup
Sign up on the Brave Search API dashboard (card required), create a key, then add npx -y @brave/brave-search-mcp-server with BRAVE_API_KEY to Claude Desktop, VS Code or fx; VS Code has one-click installs. First search within minutes.
Integrations
Eight read-only tools: web, news, image, video, local and place search, the summariser and LLM context. Goggles allow custom re-ranking, and freshness, country and safe-search filters pass straight through.
Data and privacy
The key can only spend credits and run searches; the server writes nothing. Queries go to Brave, which says they are never shared or trained on and that the API is SOC 2 Type II attested; contractual zero data retention needs Enterprise. All returned page text is untrusted input.
Cost at scale
Search: $5 per 1,000 requests at up to 50 a second. Answers: $4 per 1,000 plus $5 per million input and output tokens, 2 a second. $5 of credit per plan per month. An agent averaging three searches per task pays about $15 per 1,000 tasks.
Limits
Queries are capped at 400 characters and 50 words, and pagination stops at offset 9. No hosted endpoint, and HTTP mode is unauthenticated. The README still marks extra snippets and full local search as Pro-only, which no longer matches the prepaid plans.

What's good

  • Independent index, no Google or Bing dependency
  • llm_context returns extracted text within a set token budget
  • Eight tools across web, news, images, video, local and places
  • SOC 2 Type II; Brave says queries are not trained on

What's not

  • Card required even for the free credit
  • HTTP mode has no authentication
  • Answers plan capped at 2 requests a second
  • No Brave-hosted endpoint

Best for

Developers who want a general-purpose search tool for agents from an independent index.

Alternatives

Brave Search MCP Server: common questions

Is Brave Search MCP Server worth it?
A sound default search tool for agents, chiefly because brave_llm_context returns readable page text on the first call instead of ten links to fetch. We rate it 4.3 out of 5.
How much does Brave Search MCP Server cost?
Server is free (MIT). Brave Search API is prepaid: Search $5 per 1,000 requests (50 a second); Answers $4 per 1,000 plus $5 per million input and output tokens (2 a second). Each plan includes $5 of credit a month; card required.
What are the downsides of Brave Search MCP Server?
Card required even for the free credit; hTTP mode has no authentication; answers plan capped at 2 requests a second; no Brave-hosted endpoint.
Who is Brave Search MCP Server best for?
Developers who want a general-purpose search tool for agents from an independent index.
How much does Brave Search MCP Server cost for a team or in production?
Search: $5 per 1,000 requests at up to 50 a second. Answers: $4 per 1,000 plus $5 per million input and output tokens, 2 a second. $5 of credit per plan per month. An agent averaging three searches per task pays about $15 per 1,000 tasks.
Is Brave Search MCP Server safe for company data?
The key can only spend credits and run searches; the server writes nothing. Queries go to Brave, which says they are never shared or trained on and that the API is SOC 2 Type II attested; contractual zero data retention needs Enterprise. All returned page text is untrusted input.
What are the best Brave Search MCP Server alternatives?
The closest alternatives we have reviewed are Exa MCP Server, Tavily MCP Server, Perplexity MCP Server and Kagi MCP Server.

Answers checked Oct 22, 2026.

Make Brave Search MCP Server? Show this rating on your siteBrave Search MCP Server rated 4.3 out of 5 on ReviewRamble