gpt-agent-cli

This is a simple example of how to use the Model Context Protocol (MCP) with the OpenAI API to create a basic client that operates within a chat context.

Disclaimer

Chat messages are appended and currently the entire conversation is always sent to the server. This can rack up a lot of tokens and cost a lot of money, depending on the length of the conversation, the model you are using, and the size of the context

Limitations

This implementation currently supports only text-type tool responses.

Project Variables

Key	Description
OPENAI_API_KEY	OpenAI API Key
OPENAI_API_MODEL	Model name

Example Configuration File for Connecting MCP Servers agent-config.json

{
  "mcpServers": {
    "datetime": {
      "command": "node",
      "args": [".\\mcp-servers\\server-datetime.js"]
    },
    "systeminfo": {
      "command": "node",
      "args": [".\\mcp-servers\\server-systeminfo.js"]
    },
    "weather": {
      "command": "node",
      "args": [".\\mcp-servers\\server-weather.js"]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/User/Desktop"
      ]
    }
  }
}