Integrate Deepseek R1 for Code Generation

An MCP server implementation for the DeepSeek R1 language model, a reasoning-optimized model with an 8192-token context window, built in Node.js/TypeScript for stable MCP integration, with better type safety and error handling than the implementation team found in alternative approaches. It exposes a single deepseek_r1 tool taking a prompt plus configurable max_tokens and temperature parameters. By default it targets the deepseek-reasoner model (DeepSeek-R1); switching the model string in src/index.ts from deepseek-reasoner to deepseek-chat targets DeepSeek-V3 instead, without any other code changes required. It provides robust error handling with detailed messages covering API authentication errors, invalid parameters, rate limiting, and network issues, so failures surface with enough context to diagnose them rather than a generic error.

Use it when Claude Desktop needs to call DeepSeek R1 (or, with a one-line code change, DeepSeek-V3) for reasoning-heavy text generation. It requires Node.js 18+, npm, Claude Desktop, and a DeepSeek API key set as DEEPSEEK_API_KEY. The temperature parameter should be tuned to the task, per DeepSeek's own published guidance: 0.0 for coding and math (code generation, mathematical calculations), 1.0 for data cleaning and data analysis tasks, 1.3 for general conversation and chat dialogue, 1.3 for language translation, and 1.5 for creative writing and poetry generation - the server's own default is 0.2, distinct from all of those recommended values.

The tool's JSON call shape is fixed: {"name": "deepseek_r1", "arguments": {"prompt": "...", "max_tokens": 8192, "temperature": 0.2}} - only prompt is required; the documented example sets max_tokens to the full 8192-token ceiling and temperature to the server's 0.2 default, both of which a caller can override per request to trade off response length and creativity.

git clone https://github.com/66julienmartin/MCP-server-Deepseek_R1.git
cd deepseek-r1-mcp
npm install
cp .env.example .env  # add DEEPSEEK_API_KEY
npm run build

Claude Desktop configuration points node at the built build/index.js file with DEEPSEEK_API_KEY set in the env block. npm run dev runs the server in watch mode during development, rebuilding automatically as source files change. The project structure keeps the entire server implementation in a single src/index.ts file, compiled into build/index.js.

Developers who want Claude Desktop to call DeepSeek R1 (or DeepSeek-V3) for reasoning tasks, with task-appropriate temperature tuning and clear, diagnosable error messages when something goes wrong. Distributed under the MIT License, and open to community contributions via pull request.