Glossary
The vocabulary of mcp-pay
Payments and MCP each bring their own terms. Here is what each one means in the context of the manifest and the 402 flow.
- MCP
- The Model Context Protocol — an open protocol that lets AI models discover and call tools, resources, and prompts on external servers. mcp-pay adds payment awareness to it.
- pay.json manifest
- A static JSON document served at /.well-known/mcp/pay.json that declares per-tool pricing, accepted payment rails, and an optional payment SLA. Cacheable and readable without auth.
- HTTP 402
- The reserved "Payment Required" HTTP status code. mcp-pay returns it on paid tools instead of inventing a custom envelope, so existing clients can hook into it.
- X-PAYMENT-REQUIRED
- A response header carried on the 402 that gives an agent everything it needs to construct a payment proof for any accepted rail.
- X-PAYMENT
- A request header the agent sends on retry, carrying the constructed payment proof for the server to verify.
- Facilitator
- An out-of-band service that verifies a payment proof and confirms settlement on the underlying rail. The server asks the facilitator before running a paid tool.
- Rail
- A payment method or network — x402, MPP, Lightning, or card. mcp-pay is rail-agnostic: one manifest can advertise several rails.
- x402
- An HTTP/EVM-shaped payment protocol. It is the first rail the mcp-pay reference server implements, and one of the rails a manifest can advertise.
- MPP
- A payment rail (Tempo-network) declared as an accepted rail in the schema and planned for the reference server in a later phase.
- Lightning
- The Bitcoin Lightning Network. Expressible in the manifest via an LNURL entry in the accepts array; planned for the reference server.
- SEP-2127 (Server Card)
- The MCP Server Card proposal. server-card.json describes identity, remotes, and capabilities. mcp-pay complements it — pricing lives in pay.json, and the two link to each other.
- Payment SLA
- Optional manifest fields declaring per-rail settlement time and a refund policy, so agents can budget for latency and pick a suitable rail.