Developer Guide: Using OpenAI Models in Antigravity

This guide explains how to integrate and use OpenAI models through the Model Context Protocol (MCP) in the Antigravity environment.

1. Understanding MCP vs. Base Models

Before starting, it's important to understand how Antigravity handles models:

• Base Models (the "Brain")
  These are the primary models selected via the Model Selector in the chat interface (e.g., Gemini 3 Flash, Claude Sonnet). They act as the central reasoning engine of the agent.

• MCP Servers (the "Plugins")
  MCP servers extend the base model with tools. When you add an OpenAI MCP server, you are not adding a new selectable "brain." Instead, you are giving the currently selected base model the ability to call an OpenAI model on demand as a tool.

In short: the base model decides when to call tools; MCP servers define what tools are available.

2. Getting Your OpenAI API Key & Credits

OpenAI API usage is separate from ChatGPT Plus. To use OpenAI models via MCP, you must have API billing enabled.

2.1 Enable API Billing

• Billing dashboard:
  https://platform.openai.com/account/billing

Add a payment method and set a non-zero monthly usage limit (even $5 is enough to start).

ChatGPT Plus does not include API credits.

2.2 Create an API Key

1. Go to: https://platform.openai.com/api-keys
2. Click Create new secret key.
3. Copy the key immediately and store it securely.

3. Setting Up System Environment Variables

Storing your API key in a system environment variable is the safest and most reusable approach.

Windows (GUI)

1. Open Start Menu → search Edit the system environment variables.
2. Click Environment Variables….
3. Under User variables, click New….
4. Enter:
   • Variable name: OPENAI_API_KEY
   • Variable value: your secret API key
5. Click OK on all dialogs.
6. Restart Antigravity / your IDE.

Windows (PowerShell — optional)

setx OPENAI_API_KEY "sk-xxxxxxxxxxxxxxxx"

Restart Antigravity after running this command.

macOS & Linux

1. Open a terminal.
2. Identify your shell:

   echo $SHELL
   

3. Edit your shell config file (~/.zshrc, ~/.bashrc, etc.).
4. Add:

   export OPENAI_API_KEY='sk-xxxxxxxxxxxxxxxx'
   

5. Apply changes:

   source ~/.zshrc   # or ~/.bashrc
   

6. Restart Antigravity / your IDE.

4. Setting Up MCP Servers

Once your environment variable is set, configure the MCP server in Antigravity.

Method A: Manual Setup

1. Open Manage MCP servers in Antigravity.
2. Click Add Server.
3. Use npx to point to @pyroprompts/any-chat-completions-mcp.
4. Configure the required environment variables.

Manual setup works but can be error-prone. JSON config is recommended.

Method B: Direct JSON Config (Recommended)

1. Open Manage MCP servers.
2. Click View raw config (file icon).
3. Add the following inside the mcpServers object:

{
  "mcpServers": {
    "openai-gpt-4-1": {
      "command": "npx.cmd",
      "args": ["-y", "@pyroprompts/any-chat-completions-mcp"],
      "env": {
        "AI_CHAT_KEY": "${OPENAI_API_KEY}",
        "AI_CHAT_NAME": "OpenAI-GPT-4-1",
        "AI_CHAT_MODEL": "gpt-4.1",
        "AI_CHAT_BASE_URL": "https://api.openai.com/v1"
      }
    },
    "openai-gpt-4o-mini": {
      "command": "npx.cmd",
      "args": ["-y", "@pyroprompts/any-chat-completions-mcp"],
      "env": {
        "AI_CHAT_KEY": "${OPENAI_API_KEY}",
        "AI_CHAT_NAME": "OpenAI-GPT-4o-mini",
        "AI_CHAT_MODEL": "gpt-4o-mini",
        "AI_CHAT_BASE_URL": "https://api.openai.com/v1"
      }
    },
    "openai-gpt-5-2": {
      "command": "npx.cmd",
      "args": ["-y", "@pyroprompts/any-chat-completions-mcp"],
      "env": {
        "AI_CHAT_KEY": "${OPENAI_API_KEY}",
        "AI_CHAT_NAME": "OpenAI-GPT-5-2",
        "AI_CHAT_MODEL": "gpt-5.2",
        "AI_CHAT_BASE_URL": "https://api.openai.com/v1"
      }
    }
  }
}

Note:

• On macOS or Linux, change "command": "npx.cmd" to "command": "npx".
• Ensure Node.js is installed and available on PATH (node -v, npx -v).

5. Model Availability Notes

• Models like GPT-4.1 and GPT-4o-mini are widely available.
• GPT-5.2 requires:
  • API billing enabled
  • Explicit access to the model for your OpenAI organization

If access is missing, requests may fail with model_not_found or quota errors.

6. How to Use OpenAI Models in Chat

1. Select a Base Model (e.g., Gemini 3 Flash) from the Model Selector.
2. In chat, explicitly ask the agent to use an OpenAI tool:
   • "Use GPT-4o-mini to summarize this code."
   • "Ask GPT-5-2 to review this architecture."

The base model will decide when to invoke the tool and return the OpenAI response inside the same conversation.

Tip: Being explicit ("Use GPT-5-2…") improves tool-calling reliability.

7. Verifying Your Setup

1. Open Manage MCP servers
2. Click Refresh
3. Confirm that tools appear under each OpenAI server

If tools appear, the connection is successful.

8. Best-Practice Model Usage

Recommended setup:

| Task | Model | |-------------------------------|-------------| | Deep reasoning / architecture | GPT-5.2 | | Day-to-day coding | GPT-4.1 | | Fast or bulk tasks | GPT-4o-mini |