OpenAI Calculator

A simple calculator that uses OpenAI's function calling to process mathematical expressions and return results in JSON format.

Setup

1. Clone this repository:

   git clone https://github.com/s2005/openai-calculator.git
   cd openai-calculator

2. Set up a virtual environment:

   On Linux/macOS:

   # Create virtual environment
   python3 -m venv venv
   
   # Activate virtual environment
   source venv/bin/activate   # For bash/zsh
   # OR
   . venv/bin/activate       # Alternative method

   On Windows:

   # Create virtual environment
   python -m venv venv
   
   # Activate virtual environment
   .\venv\Scripts\activate    # For PowerShell
   # OR
   venv\Scripts\activate.bat  # For Command Prompt

3. Install the requirements:

   pip install -r requirements.txt

4. Copy .env.template to .env and add your OpenAI API key:

   cp .env.template .env
   # Edit .env and add your OpenAI API key

   Note: On Windows, you can copy the file using:

   copy .env.template .env

Usage

Basic usage:

python calculator.py "2 + 2"

With optional parameters:

python calculator.py --model gpt-4 --temperature 0.2 "23 * 45"

Command Line Arguments

• expression: The mathematical expression to evaluate (required)
• --model: OpenAI model to use (default: gpt-3.5-turbo)
• --temperature: Temperature for response generation (default: 0.0)

Example Output

{
  "expression": "2 + 2",
  "result": 4,
  "operation_type": "addition"
}

MCP Calculator Server (Anthropic Protocol)

This server implements Anthropic's Model Context Protocol (MCP) to expose the calculator functionality. It uses the mcp Python SDK (version 1.9.0 or higher).

Installation of MCP tools

The necessary MCP libraries, including the CLI, are part of the project's requirements. If you installed dependencies using pip install -r requirements.txt, the MCP tools should be available. The specific dependency is mcp[cli]>=1.9.0.

Running the Server

Ensure your OPENAI_API_KEY is correctly set in your .env file.

You can run the server using:

python mcp_calculator_server.py

Alternatively, you can use the MCP CLI for development:

mcp dev mcp_calculator_server.py

The server will register with the name "OpenAICalculatorService".

Exposed Tool

The server exposes the following tool:

• Name: evaluate_expression
• Description: Evaluates a mathematical expression using OpenAI's function calling.
• Parameters:
  • expression (string, required): The mathematical expression string to evaluate.
  • model (string, optional, default: 'gpt-3.5-turbo'): The OpenAI model to use (e.g., 'gpt-3.5-turbo', 'gpt-4').
  • temperature (float, optional, default: 0.0): The sampling temperature for generation (0.0 to 2.0).
• Returns: A dictionary containing the evaluation result. This dictionary will include fields like expression, result, operation_type, and status. If an error occurs (either in input validation, during the OpenAI call, or if the OpenAI client is not initialized), the dictionary will contain an error field and the status will be "failed".

Interaction Example

This server is designed to be used with MCP-compatible clients (e.g., an LLM that supports MCP for tool use). For development and inspection, you can explore using the mcp CLI. For instance, after starting the server, you might use commands like mcp list or mcp call (refer to MCP SDK documentation for precise usage).

A typical interaction flow would be:

1. An MCP client connects to the "OpenAICalculatorService".
2. The client invokes the evaluate_expression tool with necessary parameters (e.g., {"expression": "100 / (5 + 5)"}).
3. The server executes the tool using the process_calculation logic and returns the JSON result to the client.

Example of a successful result structure:

{
  "expression": "100 / (5 + 5)",
  "result": 10,
  "operation_type": "mixed",
  "status": "success"
}

Further Information on MCP

For more details on the Model Context Protocol (MCP) and the design philosophy behind this server's implementation, please refer to docs/llm.txt.

Error Handling

If an error occurs, the output will be in the following format:

{
  "error": "error message",
  "expression": "original expression",
  "status": "failed"
}

Requirements

• Python 3.6+
• OpenAI API key
• Required packages (see requirements.txt):
  • openai>=1.0.0
  • python-dotenv>=0.19.0
  • argparse>=1.4.0
  • mcp[cli]>=1.9.0
  • pydantic>=2.0.0