miyami-websearch-mcp

Exposes web search functionality through the Model Context Protocol (MCP) standard, allowing Claude and other MCP-compatible clients to invoke search queries as tools. Implements MCP server architecture with tool definition schemas that declare search parameters, enabling LLM agents to autonomously decide when and how to search the web without requiring direct API integration in client code.

Defines and registers web search as an MCP tool by declaring its parameters, return types, and descriptions in the MCP tool schema format. The server exposes tool metadata (name, description, input schema with JSON Schema validation) that MCP clients parse to understand how to invoke the search capability, enabling automatic tool discovery and validation without manual configuration.

Implements the MCP server lifecycle including initialization, request handling, and graceful shutdown. The server listens for MCP protocol messages (tool calls, resource requests), routes them to appropriate handlers, and returns responses in MCP format. Manages connection state and error handling to ensure reliable communication between MCP clients and the search backend.

Accepts a search query string from an MCP client, executes it against a web search backend (likely Brave Search, Google Custom Search, or similar), and formats the results into a structured response. Handles query normalization, backend API communication, result parsing, and formatting into a consistent JSON structure that includes title, URL, snippet, and metadata for each result.

Enables Claude (or other MCP-compatible LLM agents) to autonomously invoke the web search tool as part of its reasoning process. Claude's tool-use capability discovers the search tool through MCP metadata, decides when to invoke it based on context, formats the search query, and integrates results back into its reasoning loop. The agent can chain multiple searches and use results to inform subsequent actions.