Brave Search API

Executes real-time queries against a 30+ billion page index updated 100+ million times daily, returning structured results with configurable snippet counts (up to 5 per result) and schema-enriched metadata designed for RAG pipelines and LLM context windows. Results are formatted to minimize hallucination risk by providing grounded source attribution and relevance ranking optimized for AI consumption rather than human browsing.

Accepts natural language questions and returns AI-generated answers synthesized from multiple web search results, with explicit citation grounding to prevent hallucination. Implements streaming response delivery compatible with OpenAI SDK patterns, enabling real-time answer delivery to end-users. Token-based pricing tracks input and output tokens separately, allowing cost optimization for different query/answer length distributions.

Provides $5 monthly credits automatically applied to all accounts (Standard tier), enabling free experimentation and low-volume usage without upfront payment. Credits apply to both Search ($5/1000 requests) and Answers ($4/1000 requests) endpoints, providing approximately 1,000 Search requests or 1,250 Answers requests monthly at no cost. Enables developers to evaluate Brave Search before committing to paid usage.

Provides a free tier with $5 in monthly auto-credited API usage, allowing developers to experiment with Brave Search without upfront payment. The credit resets monthly and covers both Search and Answers endpoints at their respective per-request rates. Exact request quotas for the free tier are not documented, but the $5 credit translates to approximately 1,000 Search requests or 1,250 Answers requests per month.

Implements user-defined result filtering and reranking rules through the Goggles feature, allowing developers to exclude specific domains, boost results from trusted sources, or reorder results based on custom criteria. This enables application-specific search behavior without modifying the underlying query, supporting use cases like industry-specific search, content moderation, or source prioritization within RAG pipelines.

Specialized search endpoint for news content that returns recent articles with publication dates, author attribution, and source metadata. Enables temporal filtering to retrieve news from specific date ranges, supporting use cases like current events research, news aggregation, and time-sensitive RAG contexts. Results are optimized for news consumption with article-specific schema enrichment.

Dedicated image search endpoint that returns image results with URLs, alt text, source attribution, and image metadata (dimensions, file size inferred). Enables visual search integration into RAG systems and image-centric applications without requiring separate image search API. Results include source page context for understanding image provenance.

Specialized search endpoint for video content that returns video results with titles, descriptions, duration, source platform (YouTube, Vimeo, etc.), and thumbnail URLs. Enables video integration into RAG systems and multimedia applications without requiring separate video search infrastructure. Results include platform attribution and direct video links.

Implements privacy-first architecture where search queries are not profiled, stored, or sold to third parties. Enterprise tier offers explicit 'Full-funnel Zero Data Retention' guarantee, meaning no query logs, no user tracking, and no data persistence beyond immediate request processing. This enables compliance-sensitive applications (healthcare, legal, financial) to use web search without privacy concerns.

Native integration with Claude's Model Context Protocol (MCP) enabling Claude to autonomously invoke Brave Search as a tool within agentic workflows. Allows Claude to decide when to search, what to search for, and how to synthesize results without explicit developer orchestration. Implements MCP schema-based function calling with automatic tool invocation based on Claude's reasoning.

Implements separate input and output token tracking for the Answers endpoint, enabling fine-grained cost optimization based on query length and answer length independently. Developers can optimize for short queries with long answers or vice versa, and can estimate costs before making requests. Pricing is $5 per 1 million tokens (input + output combined), with token counting compatible with OpenAI's tiktoken library.

Enterprise tier provides custom rate limits, custom endpoints, dedicated support, and negotiated SLAs beyond the Standard tier's fixed 50 QPS (Search) and 2 QPS (Answers) limits. Enables high-volume deployments, mission-critical applications, and custom integration requirements. Includes invoicing, NDAs, and enterprise support channels.