Pychatter - Gemini CLI Chat Application with MCP Tools

A Python CLI application that interacts with Google's Gemini API with support for Model Context Protocol (MCP) tools. It supports both interactive chat mode and one-off commands, with the ability to call external tools for enhanced functionality.

Features

• Interactive chat mode with colored prompts and tool calling
• One-off prompt mode via command line
• MCP (Model Context Protocol) integration for external tool support
• File system operations via MCP filesystem server
• Web fetching capabilities via MCP fetch server
• Automatic tool discovery and integration with Gemini function calling
• Proper error handling with loguru
• Type-safe code with pydantic models
• Modern Python 3.12+ with type annotations
• JSON-based configuration system

Requirements

• Python 3.12+
• Node.js (for MCP servers)
• Package manager of your choice (pip, poetry, rye, etc.)

Getting started

1. Clone this repository
2. Setup config (see below).
3. Install Python dependencies:

   rye sync
   

4. Start interactive session

   rye run pychatter
   

Configuration

Create a config.json file in the project root with your settings:

{
  "gemini_api_key": "your_api_key_here",
  "gemini_api_url": "https://generativelanguage.googleapis.com/v1",
  "gemini_model": "gemini-1.5-flash",
  "log_level": "INFO",
  "mcp_servers": [
    {
      "name": "filesystem",
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/path/to/your/allowed/directory",
        "/another/allowed/directory"
      ]
    },
    {
      "name": "fetch",
      "command": "/path/to/uvx",
      "args": ["mcp-server-fetch"]
    }
  ]
}

Configuration Options

• gemini_api_key: Your Google Gemini API key (required)
• gemini_api_url: Gemini API base URL (optional, defaults to Google's API)
• gemini_model: Model to use (optional, defaults to "gemini-1.5-flash")
• log_level: Logging level (optional, defaults to "INFO")
• mcp_servers: List of MCP servers to start (optional)

MCP Server Configuration

Each MCP server in the mcp_servers array should specify:

• name: Unique identifier for the server
• command: Command to start the server
• args: Command line arguments for the server

Usage

Interactive Mode

Run in interactive mode with MCP tools:

rye run pychatter

One-off Commands

Send a single prompt:

rye run pychatter -m "Read my config.json file"

Read prompt from a file:

rye run pychatter -f path_to_file

Running Tests

rye test

License

MIT License