unichat

Unichat MCP Server is a Python-based MCP server that exposes a single tool called unichat. The tool forwards requests to a vendor or model API (OpenAI, MistralAI, Anthropic, xAI, Google AI, DeepSeek, Alibaba, Inception, etc.) via the MCP protocol. You supply a message payload to the unichat tool and receive a generated response. In addition to the unichat tool, the server provides a set of prompts that help you analyze or manipulate code, such as code_review, document_code, explain_code, and code_rework. To use the server, run it through the MCP runtime (uv/uvx) and invoke the unichat tool with the required arguments (e.g., messages) or call one of the predefined prompts with the necessary code or text payloads. You will need a vendor API key for the desired model or provider, configured via UNICHAT_API_KEY and the model selection via UNICHAT_MODEL.

Typical usage flow:

• Start the MCP server using the uvx command so it runs as a background process or in stdio mode as required by your environment.
• Use a client compatible with the MCP protocol to issue a request to the unichat tool, including the required arguments like messages (string) and optional parameters.
• If you want to analyze code, choose one of the prompts (code_review, document_code, explain_code, code_rework) and provide the code payload. The server will return the corresponding response generated by the chosen model/provider.

Prerequisites:

• Python 3.8+ installed on your system
• Git available
• MCP runtime (uv/uvx) available on your PATH

Install and run the Unichat MCP Server:

1. Clone the repository (or download the distribution):

git clone https://github.com/amidabuddha/unichat-mcp-server.git
cd unichat-mcp-server

2. Install MCP runtime (uv/uvx) if you don’t already have it. The MCP runtime is typically installed as a separate tool and provides the uv/uvx commands used to run MCP servers. For example, you can install it via your preferred method per the MCP documentation. Example (illustrative):

# Install MCP runtime if needed (adjust as appropriate for your setup)
pipx install uv

3. Configure environment variables for your provider keys and model selection. Create an environment file or export variables in your shell:

export UNICHAT_MODEL=SELECTED_UNICHAT_MODEL
export UNICHAT_API_KEY=YOUR_UNICHAT_API_KEY

4. Start the MCP server using uvx (production path):

uvx unichat-mcp-server

If you are running in development mode, you may use uv with a local directory and run command as shown in the README development snippet.

uv --directory /path/to/project run unichat-mcp-server

5. Verify the server is running and accessible via the MCP protocol according to your network configuration. You can use the MCP Inspector for debugging if needed.

Tips and notes:

• You must provide a valid UNICHAT_API_KEY for the chosen provider and set UNICHAT_MODEL to the desired model/provider combination.
• The server exposes a single tool named unichat and several prompts for code analysis and modification; use them via MCP clients that support the protocol.
• When running in development mode, you may point uv to your local source directory to streamline debugging and live edits.
• If you encounter issues with authentication, double-check API keys and ensure the provider account has access to the selected model.
• The MCP Inspector can help trace requests and debugging outputs when running the server under MCP protocol tooling.