Overview

This MCP (Model Context Protocol) server converts casual, user-friendly prompts into polished, production-ready contexts for image generation models. It sits between client applications and image-generation backends (e.g., diffusion models, hosted APIs) and performs prompt engineering, style normalization, and lightweight post-processing orchestration so models produce consistent, professional outputs with minimal client-side effort.

Designed for teams that need predictable, high-quality images at scale, the server centralizes styling presets, content policies, and transformation logic. Clients send an informal prompt and optional metadata; the server returns a refined prompt, generation parameters, and orchestration instructions that downstream model adapters can execute for fast, reliable results.

Features

Centralized prompt professionalization and normalization
Style presets (brand, photography, illustration) and parameter templates
MCP-compatible endpoints for prompt/context generation
Rate-limited, cacheable responses for improved throughput
Optional image post-processing hooks (upscale, color adjust, background removal)
Health and diagnostics endpoints for monitoring
Docker-ready, environment-driven configuration

Installation / Configuration

Prerequisites: Node.js 16+ and npm, or Docker.

Clone and run locally:

git clone https://github.com/shinpr/mcp-image.git
cd mcp-image
npm install
# run in development
npm run dev

Environment variables (example .env):

PORT=8080
MODEL_BACKEND_URL=https://api.example-model.com/v1/generate
MODEL_API_KEY=sk-xxxx
CACHE_TTL_SECONDS=300
STYLE_PRESETS_PATH=./presets.json
MAX_CONCURRENCY=8

Start with Docker:

# build
docker build -t mcp-image .
# run
docker run -p 8080:8080 \
  -e PORT=8080 \
  -e MODEL_BACKEND_URL=https://api.example-model.com/v1/generate \
  -e MODEL_API_KEY=sk-xxxx \
  mcp-image

Configuration file example (presets.json):

{
  "brand_photography": {
    "camera": "50mm, f/1.8",
    "lighting": "softbox, front",
    "color_grading": "warm, high-contrast",
    "negative_prompts": ["blurry", "low-res", "text"]
  },
  "illustration_flat": {
    "style": "flat vector",
    "palette": "brand",
    "line_weight": "2px"
  }
}

Available Resources

Endpoints

Method	Path	Description
POST	/mcp/context	Convert casual prompt + metadata into generation context and parameters
POST	/mcp/generate	Full pipeline: professionalize prompt, call model backend, return image reference
GET	/health	Service health and basic metrics
GET	/presets	List available style presets

Tools exposed by the server (logical operations available via MCP context):

professionalizePrompt: expand, disambiguate, and inject style/negative tokens
applyStylePreset: merge named preset into generation parameters
parameterTuner: adjust CFG/sampling/steps based on desired fidelity/time tradeoffs
postProcessHooks: pipeline steps for upscaling, denoising, color correction

Example request to /mcp/context:

POST /mcp/context
Content-Type: application/json

{
  "prompt": "a sleek coffee cup on a wooden table",
  "preset": "brand_photography",
  "aspect_ratio": "16:9",
  "quality": "high"
}

Example response:

{
  "refined_prompt": "A high-resolution product photo of a sleek ceramic coffee cup on a warm wooden table. 50mm lens, shallow depth of field, softbox lighting, warm color grading, high contrast, crisp detail.",
  "generation_params": {
    "width": 1920,
    "height": 1080,
    "steps": 40,
    "guidance_scale": 7.5,
    "negative_prompts": ["blur", "noise", "text"]
  },
  "post_process": ["upscale:2x", "color_grade:brand"]
}

Use Cases

E-commerce product hero images
- Input: “blue running shoes on white”
- Server output: refined prompt with studio lighting, white seamless backdrop, consistent product scale, and negative prompts to avoid text/logos; then orchestration to upscale and crop for hero banner sizes.
Marketing banners and social ads
- Input: “team celebration, happy, confetti”
- Server output: cinematic color grading, shallow depth-of-field for subject emphasis, aspect ratios for common ad slots; returns generation params tuned for fast iteration.
Product mockups for design reviews
- Input: “app UI on phone mockup”
- Server output: photorealistic phone render preset, camera angle and shadow consistency, optional PSD/transparent background post-processing.
Social media avatars and illustrations
- Input: “stylized portrait, anime”
- Server output: apply illustration_flat preset, restrict text and watermark artifacts, provide vector-friendly options and post-process for crisp line art.

Notes for Developers

The server is backend-agnostic: swap MODEL_BACKEND_URL to point to a hosted API or an internal model adapter.
Caching: commonly used prompt+preset combinations are cached to speed repeated requests.
Extendability: add presets by editing presets.json or via the /presets management endpoint.
Security: enforce content policy checks before sending prompts to model backend; configure API keys through environment variables.

This MCP server is intended as a middleware layer that standardizes prompt quality across teams and production pipelines, reducing variance and improving visual consistency for downstream image generators.

Intelligent Image Generator MCP Server for Professional Images