ChatGPT GPT Manager

An MCP (Model Context Protocol) server for creating, managing, and testing custom GPTs on ChatGPT.

Features

• Initialize browser sessions for ChatGPT
• Create new custom GPTs with specified names and instructions
• Test custom GPTs with specific prompts
• Capture screenshots of the process
• Work with multiple GPTs simultaneously

Installation

Using NPM

npm install -g chatgpt-gpt-manager

Using Docker

docker pull samihalawa/chatgpt-gpt-manager
docker run -p 8080:8080 -v $(pwd)/temp:/app/temp samihalawa/chatgpt-gpt-manager

Using Smithery

npx @smithery/cli run chatgpt-gpt-manager

Usage

Starting the Server

npx chatgpt-gpt-manager

The server will start and be available on the specified port (default is 8080).

MCP Functions

The server provides the following MCP functions:

browser_initialize

Initialize a browser session for ChatGPT automation.

{
  "function": "browser_initialize",
  "arguments": {}
}

create_gpt

Create a new custom GPT on ChatGPT.

{
  "function": "create_gpt",
  "arguments": {
    "name": "My Test GPT",
    "instructions": "You are a helpful assistant that provides information about planets."
  }
}

Optional parameters:

• options: Additional options for GPT creation

test_gpt

Test a custom GPT with a specific prompt.

{
  "function": "test_gpt",
  "arguments": {
    "gptId": "g-abc123def456",
    "prompt": "Tell me about Mars."
  }
}

browser_close

Close the browser and release resources.

{
  "function": "browser_close",
  "arguments": {}
}

Configuration

The server accepts the following environment variables:

• HEADLESS: Set to “true” to run the browser in headless mode (default: false)
• DEBUG: Set to “true” to enable debug logging (default: false)
• SCREENSHOT_DIR: Directory to store screenshots (default: “./temp”)
• MAX_CONCURRENT_SESSIONS: Maximum number of concurrent browser sessions (default: 3)

Important Notes

• This tool requires you to be logged in to ChatGPT. If not logged in, the browser window will open for manual login.
• Custom GPT creation process might change as OpenAI updates the ChatGPT interface.
• It is recommended to run with HEADLESS=false initially to see the process and ensure everything works correctly.

License

MIT