The Vectara MCP Server implements the Model Context Protocol (MCP) as a secure gateway to Vectara's Trusted RAG platform. It enables AI agents and applications to interact with data sources and tools through MCP, delivering fast, reliable search and generation with reduced hallucination. Compatible with Claude Desktop and any MCP client, it supports multiple transport modes (HTTP/SSE and STDIO) to cover production deployments and local development. By default, the HTTP/SSE transport is secured with bearer token authentication and can be configured to enforce TLS, CORS, and origin restrictions. STDIO mode offers low-latency local communication suitable for desktop workflows, but is explicitly less secure and must be enabled with the --stdio flag. Configuration is flexible via command-line options (--host, --port, --transport, --path, --no-auth) and environment variables (VECTARA_API_KEY, VECTARA_AUTHORIZED_TOKENS, VECTARA_ALLOWED_ORIGINS, VECTARA_TRANSPORT, VECTARA_AUTH_REQUIRED). The server exposes a set of tools (ask_vectara, search_vectara, correct_hallucinations, eval_factual_consistency) and API-key management utilities (setup_vectara_api_key, clear_vectara_api_key) for end-to-end workflows. Claude Desktop integration provides sample configurations for STDIO transport.