Imagegen MCP

What imagegen-mcp MCP server does

The imagegen-mcp MCP server wraps OpenAI’s image generation and editing APIs and exposes them as MCP tools. It supports text-to-image creation with models listed by the project, including gpt-image-1, dall-e-2, and dall-e-3 when available and enabled. It also supports image editing with gpt-image-1 and dall-e-2.

The server provides two tools:

• text-to-image creates an image from a required text prompt.
• image-to-image edits one or more local image files using a required prompt and an optional PNG mask.

Results are written to temporary files. The tool response includes the path to the generated image, allowing the MCP client or agent to locate the output locally.

How it works

The imagegen-mcp MCP server runs locally and communicates with an MCP client through standard input and output using StdioServerTransport. It sends image requests to OpenAI with the selected model and request parameters, then saves the returned image data as a temporary file.

For text generation, optional settings include image size, output format, compression, moderation, background, quality, style, and image count. The available values depend partly on the selected OpenAI model. For example, style applies to DALL·E 3, while transparent backgrounds require PNG or WebP output. DALL·E 3 supports only one image per request.

Editing accepts local file paths rather than remote image URLs. An optional mask file must be a PNG; transparent areas identify the parts of the image to edit.

Setup and configuration

The project requires Node.js, with version 18 or later recommended, npm or yarn, and an OpenAI API key. Set the key in the OPENAI_API_KEY environment variable. A cloned installation can use the repository’s .env.example as the starting point for a .env file.

You can run the package without cloning the repository:

npx imagegen-mcp [options]

For a local checkout, install dependencies, build the TypeScript source, and run either the development entry point with ts-node or the compiled JavaScript in dist. The --models option restricts which OpenAI models the server exposes. Without it, the default allowlist contains gpt-image-1, dall-e-2, and `dall-e-3 as defined by the project.

In Cursor, add the server as a custom MCP server and provide OPENAI_API_KEY in its environment configuration. The command can include --models to limit the models available to Cursor.

Tools and capabilities

text-to-image accepts a required text value and optional model, size, style, output format, compression, moderation, background, quality, and n parameters. Its default image size is 1024x1024, and the default output format is PNG.

image-to-image accepts local image paths, an edit prompt, and an optional mask. It supports output format, compression, quality, size, model, and count options. Editing is limited to gpt-image-1 and dall-e-2; DALL·E 2 supports only its documented image sizes.

The imagegen-mcp MCP server also supports npm-based execution, local development, TypeScript compilation, linting, and formatting commands described by the project.

Limitations and notes

Model access depends on the OpenAI API key and the models enabled for that account. Model-specific restrictions apply to sizes, styles, quality settings, transparency, and image counts. The server uses local temporary files, so callers must be able to access the returned paths. Image editing requires local input paths and does not accept URLs according to the documented tool parameters.