A minimal Model Context Protocol (MCP) server that exposes a single tool, generate, which creates an image from a text prompt and returns it as a base64-encoded PNG MCP resource. The server demonstrates how to make a paid MCP tool using paymcp with Stripe or Walleot.

• How to wire paymcp into an MCP server and set a price for a tool.
• A minimal shape for a paid, single-purpose MCP tool you can extend.
• How to serve an MCP server over Streamable HTTP with Express (POST/GET/DELETE to /mcp).

• Node.js 18+
• OpenAI API key: set OPENAI_API_KEY
• Stripe secret key (optional): set STRIPE_SECRET_KEY if you want to enable Stripe payments (test keys are fine for development)
• Walleot API key (optional): set WALLEOT_API_KEY and uncomment the provider in code if you want to try Walleot instead of Stripe

# clone the repo
pnpm install

This project is TypeScript. Use your preferred runner (e.g., tsx, ts-node) when bootstrapping the server process if you create a custom entry point.

Set environment variables (for example via a .env that your runner loads, or via your shell):

export OPENAI_API_KEY="sk-..."
# Optional, if using Stripe
export STRIPE_SECRET_KEY="sk_test_..."
# Optional, if using Walleot (then uncomment provider in the server code)
export WALLEOT_API_KEY="..."

Payment providers are configured in installPayMCP within the server code:

installPayMCP(server, {
  providers: {
    // "walleot": { apiKey: process.env.WALLEOT_API_KEY ?? "" }, // Uncomment to enable Walleot
    "stripe": { apiKey: process.env.STRIPE_SECRET_KEY ?? "" },
  },
  mode: mode.ELICITATION,
});

Payment notes: Stripe has a minimum charge per transaction. If your tool price is below $2.00, prefer Walleot; small Stripe payments may fail or be uneconomical.

• The server registers a single tool generate:
  • Input: { prompt: string }
  • Price: $2.00 per call (configured in the code; change as you like)
  • Output: MCP resource of type image with base64-encoded PNG data
• Internally it calls generateImage(prompt) from src/services/openai.ts, which uses the official OpenAI SDK, requests an image, and returns a base64 string.

The tool returns a response like:

{
  "content": [
    {
      "type": "image",
      "data": "<base64>",
      "mimeType": "image/png",
      "annotations": {
        "audience": ["user"],
        "priority": 0.9
      }
    }
  ]
}

This server is already wired to run over HTTP using Express and the StreamableHTTPServerTransport.

# dev with tsx watch
pnpm dev
# build TypeScript -> build/
pnpm build
# start compiled server from build/
pnpm start

By default it listens on PORT (defaults to 5004):

https://localhost:5004/mcp

• MCP Inspector:

  1. Start this server (pnpm dev) so it listens on https://localhost:5004/mcp.
  2. In a new terminal, run:

     npx @modelcontextprotocol/inspector@latest

  3. Open the URL printed by Inspector (e.g., https://localhost:5173).
  4. Click Add server → select Streamable HTTP → set URL to https://localhost:5004/mcp → Connect.
  5. In the tools list, choose generate, provide a prompt, and run it to receive a base64 PNG.

• Demo client: Try the companion chat client here: https://github.com/Walleot/walleot-demo-chat

• Change price: in server.registerTool update price: { amount: 2, currency: "USD" }. If you use Stripe, set the price above Stripe’s minimum; for prices below $2.00, prefer Walleot.
• Change output: adjust src/services/openai.ts if you want URLs instead of base64, larger sizes, different formats, etc. The current function is intentionally minimal and always returns a base64 PNG string.
• Swap payment provider: switch from Stripe to Walleot by toggling the provider block and keys.

MIT

Notes on package manager: this repo declares packageManager: pnpm. Examples above use pnpm. If you prefer others, use: npm run dev|build|start or yarn dev|build|start.