Monetize your AI & ML MCP server compute-backed calls deserve compute-backed prices.

A directory lookup can plausibly be free forever; a model inference cannot. Servers like MinerU Open MCP (document parsing across PDF, DOCX, PPTX, and more) and selvage (AST-aware LLM code review) sit on top of upstream APIs and GPU time that someone is already paying for. Per-call pricing simply moves that cost — plus a margin — to the agent that benefits.

The buyer side is equally aligned. An agent mid-task doesn't want a subscription to a PDF parser it needs once; it wants to pay for one parse and move on. Pay-per-call is the shape of demand here, not just the shape of supply.

The unit of value varies wildly inside this category, so anchor on what one call actually does. Lightweight calls — a time-mcp timezone lookup, an anki-mcp-server flashcard query — belong at Loomal's $0.01 minimum. They cost you almost nothing and the floor keeps them sustainable.

Heavy calls deserve heavy prices. A full document parse, a multi-file code review, or a governed multi-step evaluation can reasonably run from a few cents to over a dollar per call, because each one consumes seconds of GPU or thousands of model tokens. A useful sanity check: price each tool at your marginal compute cost times two to three, then adjust against what the output would cost the caller to produce themselves.

Loomal's index has 96 live AI & ML servers, and most are unclaimed. If yours is among them, claim it, prove ownership through GitHub, and the listing becomes yours to price. Per-call pricing is one field in the console; changing it later is the same field.

Deploy the x402 middleware in front of your remote endpoint and the flow is automatic: unpaid calls get an HTTP 402 with your price, paid calls execute and return an Ed25519-signed receipt. Loomal takes 5% of settled transactions — a fee that's currently waived.

Because x402 prices attach per tool call rather than per account, you can run differentiated pricing inside one server. An Apollo MCP Server-style deployment exposing GraphQL operations could keep schema introspection at $0.01 while pricing mutation-executing operations higher. No plans, no seats, no quota tables — the price sits on the call itself.

That granularity also makes experimentation cheap: nudge one tool's price, watch call volume in the console, and converge on what the market of agents will actually pay.