Cloudflare Image MCP

OpenAI-compatible image generation API + Streamable HTTP MCP server powered by Cloudflare Workers AI.

🌟 Features

• OpenAI-Compatible API: /v1/images/generations + /v1/images/edits endpoints
• MCP Protocol Support: HTTP transport with SSE (streamable)
• 10 Image Generation Models: FLUX, SDXL, Stable Diffusion, and more
• Multiple Tasks: Text-to-image, image-to-image (masked edits supported)
• Web Frontend: React SPA for image generation and provider credentials
• R2 Storage: Auto-expiring image storage with CDN

📦 What’s Included

cloudflare-image-mcp/
├── workers/           # Cloudflare Worker (API + MCP)
├── client/            # React SPA (Vite + Tailwind + TanStack Query)
└── e2e/               # Playwright E2E tests

---

🚀 Quick Start

Prerequisites

1. Deploy this MCP server to your Cloudflare account (takes 5 minutes): 👉 Deployment Guide

2. Note your worker URL after deployment:

   https://cloudflare-image-workers..workers.dev
   

3. If you set API_KEYS during deployment, you’ll need the key for authentication.

MCP Client Configuration

Below are the configuration guides for different MCP clients. This server uses HTTP/SSE transport (not stdio) since it runs on Cloudflare Workers.

---

📖 Usage

OpenAI-Compatible API

curl -X POST "https://your-worker.workers.dev/v1/images/generations" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-api-key" \
  -d '{
    "model": "@cf/black-forest-labs/flux-1-schnell",
    "prompt": "A beautiful sunset over mountains",
    "n": 1,
    "size": "1024x1024"
  }'

And it can be used for OpenWebUI’s image generation feature and image edits feature:

MCP Tools

Available Tools:

• list_models - List all available models
• describe_model - Get model parameters and limits
• run_model - Generate images

Connect via MCP:

{
  "mcpServers": {
    "image-gen": {
      "url": "https://your-worker.workers.dev/mcp"
    }
  }
}

Web Frontend

Open your worker URL in a browser for an interactive UI. The frontend is a React single-page app deployed as Workers Static Assets: the Worker owns the API routes and hands the client’s build output to the browser. /admin opens the provider credentials workbench.

---

🎨 Supported Models

📋 Detailed Model Specifications → — Full parameter reference, feature comparison, and capability matrix.

Provider = who built the model. Hosted By = who actually serves the inference API (and bills for it): Cloudflare Workers AI for the @cf/* models, Agnes AI’s own OpenAI-compatible endpoint for agnes-image-*. Cloudflare’s ids are namespaced (@cf/{provider}/{model}); the Agnes id is the bare upstream model name, which is what the transport sends.

---

📚 Documentation

---

🧪 Testing

# Run E2E tests against staging/production
npm run test:e2e:staging
npm run test:e2e:production

---

🔧 Development

# Type check
npm run build

# Run E2E tests
npm run test:e2e:staging

---

📝 License

MIT

---

Built with:

• Cloudflare Workers AI
• Cloudflare R2
• Model Context Protocol