Server payments

Server payments

Server payments are implemented as middleware that sits between:

• the inbound client request, and
• the underlying MCP server handler.

For priced requests, the middleware ensures no unpaid forwarding.

Priced capability matching

You price individual capabilities by method + name.

import type { PricedCapability } from '@contextvm/sdk/payments';

const pricedCapabilities: PricedCapability[] = [
  { method: 'tools/call', name: 'add', amount: 10, currencyUnit: 'sats' },
  {
    method: 'resources/read',
    name: 'private://*',
    amount: 5,
    currencyUnit: 'sats',
  },
];

Notes:

• Only requests that match a priced capability are gated.
• Matching is designed for predictable policy. Prefer explicit entries over overly-broad wildcards.

Processors (server-side payment rails)

A PaymentProcessor is responsible for:

• creating a pay_req for a given amount
• verifying that a previously issued pay_req has been paid

You can configure multiple processors (multiple PMIs). The server selects a processor based on client/server PMI compatibility.

import {
  LnBolt11NwcPaymentProcessor,
  withServerPayments,
} from '@contextvm/sdk/payments';

const processor = new LnBolt11NwcPaymentProcessor({
  nwcConnectionString: process.env.NWC_SERVER_CONNECTION!,
});

withServerPayments(transport, {
  processors: [processor],
  pricedCapabilities,
});

Dynamic pricing: resolvePrice

resolvePrice runs on every priced request and returns the final quote.

import type { ResolvePriceFn } from '@contextvm/sdk/payments';

const resolvePrice: ResolvePriceFn = async ({
  capability,
  request,
  clientPubkey,
}) => {
  // Example: price based on request size.
  const requestSize = JSON.stringify(request.params ?? {}).length;
  const extra = Math.ceil(requestSize / 1024);
  const amount = Math.max(1, Math.round(capability.amount + extra));

  return {
    amount,
    description: `Request size: ${requestSize} bytes`,
    _meta: { requestSize },
  };
};

withServerPayments(transport, {
  processors: [processor],
  pricedCapabilities,
  resolvePrice,
});

Guidance:

• Keep it fast and deterministic.
• Treat it like authorization + pricing logic.
• If you run multiple server instances, store any usage/quota state in a durable store.

Rejecting requests without charging

To reject a priced request without creating an invoice, return { reject: true, message? } from resolvePrice.

import type { ResolvePriceFn } from '@contextvm/sdk/payments';

const resolvePrice: ResolvePriceFn = async ({ capability, clientPubkey }) => {
  const isBlocked = await isUserBlocked(clientPubkey);
  if (isBlocked) {
    return { reject: true, message: 'Access denied' };
  }

  return { amount: capability.amount };
};

When rejected:

• the server emits notifications/payment_rejected correlated to the request
• no processor method is called
• the request is not forwarded

Notifications and correlation

Payment notifications are correlated to the original request using an e tag (the request event id).

• notifications/payment_required
• notifications/payment_accepted
• notifications/payment_rejected

This is how clients know which in-flight request a payment notification belongs to.