A Model Context Protocol (MCP) server for the DeepSeek API, enabling seamless integration of DeepSeek's advanced language models with MCP-compatible applications such as Claude Desktop. The server supports anonymous usage by routing requests through a proxy, so only a proxy is seen on the far side. It can be installed via Smithery or npm and configured for Claude Desktop, including an example claude_desktop_config.json snippet. It exposes resources to discover available models and configurations, and to retrieve or modify settings such as custom model selection, temperature, max tokens, Top P, presence penalty, and frequency penalty. Automatic model fallback is provided: if the primary model (deepseek-reasoner / R1) is down, the server automatically falls back to the secondary model (deepseek-chat / V3), with easy switch prompts to alternate between them. Enhanced multi-turn conversation support preserves complete history and settings for long-running interactions, training/fine-tuning data generation, and complex dialogues. Local testing is supported via MCP Inspector (build, run, test chat completions, debug, monitor). By default, the server uses the DeepSeek R1 model for reasoning and general tasks while providing access to model/config resources.