# OpenAI GPT MCP Server

A local [Model Context Protocol](https://modelcontextprotocol.io/) server that provides OpenAI GPT text generation over standard input/output.

## Prerequisites

- Node.js 18 or newer
- An OpenAI API key

## Install and build

```powershell
npm install
npm run build
```

Copy `.env.example` to `.env` and set your API key:

```powershell
OPENAI_API_KEY=your_api_key
```

`OPENAI_MODEL` is optional and defaults to `gpt-5`. `OPENAI_IMAGE_MODEL` is optional and defaults to `gpt-image-2`.

## MCP client configuration

Build first, then add this server to your MCP client's configuration. Replace `<absolute-path-to-repository>` with the absolute path to your clone.

```json
{
  "mcpServers": {
    "openai-gpt": {
      "command": "node",
      "args": ["<absolute-path-to-repository>\\dist\\index.js"]
    }
  }
}
```

## Tool: `generate_text`

Required input:

- `prompt`: text to send to the model.

Optional inputs:

- `instructions`: system-level behavior for the response.
- `model`: a GPT model ID, overriding `OPENAI_MODEL`.
- `max_output_tokens`: maximum response size, from 1 through 16,384.

The server uses the OpenAI Responses API and returns generated text to the MCP client. It never writes API keys to output or logs.

## Tool: `generate_image`

Generates one PNG image and returns it directly to the MCP client.

- `prompt` (required): detailed image description.
- `model` (optional): image model ID, overriding `OPENAI_IMAGE_MODEL`.
- `size` (optional): `1024x1024`, `1536x1024`, or `1024x1536`.
- `quality` (optional): `low`, `medium`, or `high`.

Image generation consumes OpenAI API credits separately from Claude usage.

## Run directly

```powershell
npm start
```

For interactive inspection:

```powershell
npm run inspect
```