gemini-image-studio-mcp

MCP server for AI image generation and editing with Google Gemini. Create web assets, ad creatives, and brand visuals — with structured JSON editing for precise, repeatable control.

What Makes This Different

Most Gemini image MCP servers are basic text-to-image wrappers. This one adds a structured editing pipeline:

1. Generate an image from text or JSON prompts

2. Decompose it into a structured JSON blueprint (every visual component mapped)

3. Edit by changing specific fields — subject[0].hair.color: "platinum_blonde" — and regenerating

This means precise, isolated changes without affecting the rest of the image. Change a hair color without touching the background. Swap clothing without altering the pose. All through dot-notation JSON paths.

Related MCP server: universal-image-mcp

Features

• 5 MCP Tools — generate, decompose, edit, presets, list

• Structured JSON Editing — decompose images into blueprints, edit specific fields with dot-notation

• Natural Language Editing — or just describe the change in plain English

• 10 Built-in Presets — Facebook ads, Instagram stories, hero images, OG images, YouTube thumbnails, and more

• Reference Image Support — up to 14 reference images for character/object consistency

• Dual Model Support — Gemini 3.1 Flash (fast) or Gemini 3 Pro (best quality)

• Blueprint Caching — decomposed blueprints cached alongside images for instant re-edits

• Google Search Grounding — real-world accuracy via web search

• Smart Error Handling — retry on rate limits, clear safety block messages, file size warnings

Quick Start

1. Get a Gemini API Key

Get one free at Google AI Studio.

2. Install

npm install -g gemini-image-studio-mcp

3. Add to Claude Code

claude mcp add gemini-image-studio-mcp -e GEMINI_API_KEY=your-key-here -- gemini-image-studio-mcp

Or add to your project's .claude/mcp.json:

{
  "mcpServers": {
    "gemini-image-studio-mcp": {
      "command": "npx",
      "args": ["-y", "gemini-image-studio-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-key-here"
      }
    }
  }
}

4. Use It

Ask Claude to generate images:

"Create a Facebook ad for a coffee shop with warm lighting"

"Generate a hero image for a tech startup landing page"

"Edit the hero image — change the background to a sunset beach"

Tools

generate_image

Create a new image from text or structured JSON prompts.

decompose_image

Analyze an image into a structured JSON blueprint — the first step of the edit workflow.

Returns a full blueprint with subject, scene, technical, composition, text_rendering, style_modifiers, and meta sections — each field precisely describing the image's visual components.

edit_image

Edit an image using JSON changes or natural language.

JSON edit example — change hair color and add sunglasses:

{
  "image_path": "/output/portrait.png",
  "edit_type": "json",
  "changes": {
    "subject[0].hair.color": "platinum_blonde",
    "subject[0].accessories": [
      { "item": "sunglasses", "material": "metal", "color": "#C0C0C0" }
    ]
  }
}

Natural language edit example:

{
  "image_path": "/output/portrait.png",
  "edit_type": "natural_language",
  "instruction": "Change the background to a tropical beach at sunset. Keep the person exactly the same."
}

get_presets

List available asset presets with dimensions, tips, and conventions.

list_generated

Browse previously generated images.

JSON Editing Workflow

The key differentiator — precise, field-level image editing:

Step 1: Generate
  generate_image(prompt: "Professional headshot, navy blazer", preset: "linkedin_post")
  → /output/headshot.png

Step 2: Decompose
  decompose_image(image_path: "/output/headshot.png")
  → JSON blueprint with every visual component mapped

Step 3: Edit (precise)
  edit_image(
    image_path: "/output/headshot.png",
    edit_type: "json",
    changes: {
      "subject[0].clothing[0].color": "#8B0000",
      "scene.lighting.type": "studio_softbox"
    }
  )
  → /output/headshot-edit-1.png (blazer changed to dark red, lighting adjusted)

Step 4: Edit (creative)
  edit_image(
    image_path: "/output/headshot-edit-1.png",
    edit_type: "natural_language",
    instruction: "Add warm bokeh to the background"
  )
  → /output/headshot-edit-1-edit-1.png

Dot-Notation Paths

subject[0].hair.color          → Hair color
subject[0].hair.style          → Hair style
subject[0].clothing[0].color   → First clothing item color
subject[0].accessories         → Add/change accessories
scene.lighting.type            → Lighting type
scene.location                 → Location/background
text_rendering.text_content    → Text in image
technical.lens                 → Camera lens
composition.framing            → Shot framing
style_modifiers.aesthetic      → Aesthetic style

Built-in Presets

Configuration

Integration

Claude Code

claude mcp add gemini-image-studio-mcp -e GEMINI_API_KEY=your-key -- gemini-image-studio-mcp

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "gemini-image-studio-mcp": {
      "command": "npx",
      "args": ["-y", "gemini-image-studio-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-key-here"
      }
    }
  }
}

Any MCP Client

GEMINI_API_KEY=your-key npx gemini-image-studio-mcp

The server communicates over stdio using the Model Context Protocol.

MCP Prompt & Resource

This server also exposes:

• Prompt: nano_banana_expert — invoke this to give Claude full knowledge of the JSON schema, editing best practices, and asset creation guidelines

• Resource: nanobanana://schema/prompt — the raw JSON schema with all enum values for programmatic access

Models

Contributing

Contributions welcome! Please:

1. Fork the repository

2. Create a feature branch (git checkout -b feature/my-feature)

3. Run tests (npm test)

4. Commit your changes

5. Push and open a PR

License

MIT