What people and their agents pay for: x402, Stripe and marketplace data by community

Our last post counted the MCP servers, connectors and skills each community installs. Installing is free. This one follows the money: what each community pays for, and what agents now pay for themselves.

No large tool maker publishes its revenue, and Stripe publishes no figure for money paid to MCP servers. So we read the payments that are public: every x402 payment on chain, Apify’s marketplace, and the revenue small makers verify through Stripe, Polar or Paddle. The short answer: people pay most for data about people (social profiles, leads, places), agents paying for themselves buy model calls and API access, and the communities paying are marketers, sales and creators. Developers install the most, and pay mainly for model calls.

Stripe carried $1.9 trillion in 2025 and says the top AI companies on it reach $1 million in revenue about four months sooner than the top software companies did. It launched payments for agents (machine payments, an agentic commerce suite, a wallet agents can use), but has not said how much flows through them.

Agent tools are paid for three ways: per request, by the agent itself, over x402; per result, through a marketplace like Apify; and by subscription, the way any software is. The three report different things, so the figures below are not a ranking.

x402 lets an agent pay for an API call itself, in stablecoins, with no account. In the last 30 days, 36,563 buyers paid 39,175 sellers $1,090,730 in 2,831,358 payments, well under a dollar each.

By count, agents buy model calls and API gateways. By dollars, the top seller is not an agent tool at all: people turning stablecoins into prepaid cards. After that come crypto markets and voice. Sales and marketing data, the leads and social profiles that dominate Apify, is still small here.

Apify sells scrapers and agents by the result, and its store says how many people used each in the last 30 days. Social media and lead generation lead by far: marketers watching Instagram and TikTok, sales teams building lists from Google Maps and LinkedIn. Apify says it paid its developers $1.7 million last month. Of its 852 most popular actors, pay-per-result ones had 27 times the users of free ones.

TrustMRR lists makers who let it read their revenue straight from their payment provider. It is not a census, since makers choose to list, but every figure is real money. Grouped by who buys:

One model gateway for developers earns the most. After it, the tools earning most get data or results for marketers, creators and sales, and other developer tools earn when they cut model bills. Selling skills earns little: of four skill directories listed, one had revenue in the last 30 days. Everyday users pay to have OpenClaw set up for them: TrustMRR’s OpenClaw category holds 187 startups and $156,718 in the last 30 days.

Most of the tools in our last post are free to install and paid to use: the agent needs the plan behind the server. These are the cheapest plans that let an agent use each one.

People spend on the AI apps themselves. Sensor Tower puts in-app purchases in AI apps on course to pass $4 billion in the first half of 2026, up 36% on the half before, and Claude’s average revenue per US mobile user rose from under $0.50 to $2.76 between September 2025 and May 2026. Nobody breaks that spending down by what the agent is connected to.

Large makers publish nothing. Firecrawl, Exa, Browserbase, Apify, n8n and ElevenLabs do not show verified revenue anywhere public, so the subscriptions table is small makers only.

x402 totals in the press are inflated. Articles cite 138 million and 200 million payments; a Visa and Artemis analysis put real volume at about $19 million once wash trading is removed. We use x402scan’s last 30 days and leave out sellers whose payments look automated, such as one with 470,969 payments from 151 buyers for $1,071.

Card networks publish pilots, not volumes. Visa, Mastercard, PayPal and Shopify have announced agent checkout, and none has said how much has gone through it.

hedwigai’s skills answer from public records, the way this post was made: web traffic and rankings, leads and hiring, SEC and FDA records, and a reference checker. Reading costs nothing; metered data is billed per call.

npx skills add un-seen/hedwigai-skills