Brave Search MCP Server

Executes web searches through the Brave Search API and exposes results as MCP tools that LLM clients can invoke. Implements the MCP tool-calling protocol to register a 'search' tool with schema validation, marshals user queries to Brave's REST API, and returns structured results (title, description, URL, favicon) that the LLM can reason over and cite. Uses JSON-RPC 2.0 message passing over stdio or HTTP transport to communicate between client and server.

Performs location-based business searches through Brave Search's local business API, returning structured results with business names, addresses, phone numbers, and ratings. Implements MCP tool registration for local search with geographic parameters, translates LLM-friendly queries into Brave's business search schema, and returns results formatted for LLM consumption. Operates through the same MCP JSON-RPC transport as web search but routes to Brave's specialized local business endpoint.

Registers search capabilities as MCP tools with full JSON schema definitions, enabling clients to discover tool signatures, parameter types, and descriptions before invocation. Implements MCP's tool registration protocol by defining tool metadata (name, description, input schema) and validating incoming requests against the schema before forwarding to Brave API. Uses TypeScript SDK's tool builder to construct schemas that LLMs can parse to understand what parameters are required, optional, and what types they accept.

Provides transport-agnostic MCP server implementation that can communicate with clients via stdio (for local/CLI integration) or HTTP (for remote/server deployment). The MCP TypeScript SDK abstracts transport details, allowing the Brave Search server to register tools once and work with any MCP-compatible client regardless of transport. Handles JSON-RPC 2.0 message framing, request/response correlation, and error propagation across both transports without server-side code changes.

Manages Brave Search API authentication by reading API keys from environment variables and passing credentials to Brave endpoints on each request. Implements credential handling through standard Node.js environment variable patterns (BRAVE_API_KEY), avoiding hardcoded secrets in code. Forwards credentials to Brave's REST API in request headers, handling authentication errors and rate-limit responses from Brave's servers.

Transforms Brave Search API responses into structured JSON that LLMs can easily parse and reason about. Extracts relevant fields (title, description, URL, favicon) from Brave's raw API response, removes HTML/markup, and formats results as a clean array that the LLM can iterate over and cite. Implements result normalization to handle missing fields, truncate long descriptions, and ensure consistent output structure across different search types (web vs local business).

Catches errors from Brave Search API calls (rate limits, invalid keys, network failures) and translates them into MCP-compatible error responses that clients can handle gracefully. Implements error classification (authentication errors, rate limits, network errors) and returns structured error messages with HTTP status codes and Brave API error details. Prevents server crashes by wrapping API calls in try-catch blocks and returning error responses instead of throwing exceptions.

Serves as an official reference implementation of MCP server patterns using the TypeScript SDK, demonstrating best practices for tool registration, request handling, and transport abstraction. Provides example code that developers can study and adapt for building their own MCP servers. Maintained by Anthropic's MCP team as part of the official MCP ecosystem, ensuring patterns align with protocol evolution and SDK updates.