The point is provider-independence: the same unichat tool fronts OpenAI, Anthropic or any compatible endpoint, so getting a different model's take is a config change rather than a different server. The four prompts are the practical add-on — code review, documentation, explanation and rework as ready-made instructions rather than tools.
A server that exposes a single tool for chatting with a configured model, plus a set of code-focused prompts, across many providers.
- unichat — one tool that sends messages to the configured model and returns a response
- A code_review prompt that reviews code for issues and improvements
- A document_code prompt that generates docstrings and comments
- An explain_code prompt that walks through how a piece of code works
- A code_rework prompt that applies requested changes to code
- One configuration reaches OpenAI, Anthropic, and OpenAI-compatible providers such as MistralAI, xAI, Google AI, DeepSeek and Alibaba
A vendor API key. Set UNICHAT_MODEL to the model and UNICHAT_API_KEY to its key; for an OpenAI-compatible provider, add UNICHAT_BASE_URL pointing at its endpoint, and the built-in model list is skipped. Run it with uvx and the unichat-mcp-server package.
One command plus a key — npx -y @smithery/cli install unichat-mcp-server --client claude, then supply credentials