Bryce Watson
Back to projects
•
3 min read
Stylize MCP Server

A production-deployed MCP server that brings AI image stylization into Claude Desktop conversations. Upload or describe an image, pick a style, and get a styled result back — without leaving the chat.

Problem

AI assistants can discuss images, analyze them, and reason about visual concepts, but they can’t generate or transform images natively within a conversation. Users have to context-switch to separate tools for anything visual, breaking their workflow.

Approach

Stylize MCP Server exposes image stylization as an MCP tool that Claude Desktop can call directly via SSE transport. The server handles the full pipeline: GPT-4V analyzes the input image for context, a template-driven prompt system constructs the style request, and DALL-E 3 generates the output. Results are cached in Redis and delivered through Google Cloud Storage.

The service runs on GCP Cloud Run with a freemium credit system across four pricing tiers. Abuse prevention includes device fingerprinting, VPN detection, and behavioral analysis. Content safety filtering is handled by Cloud Vision SafeSearch before any image is returned.

What’s Implemented

  • Five artistic styles — Van Gogh, Pixel Art, Flat UI, Neumorphic, and Glassmorphic, each with tuned prompt templates
  • MCP integration — SSE transport for real-time Claude Desktop communication, following the Model Context Protocol specification
  • Smart image pipeline — GPT-4V context analysis feeds into style-specific prompt construction for higher-quality results
  • Web demo interface — Drag-and-drop upload with style selection, available without signup (5 free images)
  • SaaS infrastructure — Freemium credit system, four pricing tiers, Redis caching, GCS asset delivery

Architecture

Claude Desktop ──SSE──▸ MCP Server (FastAPI / Cloud Run)
                            │
                  ┌─────────┼─────────┐
                  ▼         ▼         ▼
              GPT-4V    DALL-E 3   Cloud Vision
            (analyze)  (generate)   (safety)
                  │         │
                  ▼         ▼
               Prompt    Redis ──▸ GCS
             Templates   Cache    Storage

What This Demonstrates

End-to-end MCP protocol implementation in a production setting, including SSE transport, tool registration, and real-time response streaming. Also demonstrates SaaS product design (freemium monetization, abuse prevention, content safety), cloud-native deployment on GCP, and integration of multiple AI services (GPT-4V, DALL-E 3, Cloud Vision) into a single coherent pipeline.