This preprint presents a taxonomy of payment coordination patterns for MCP (Model Context Protocol) servers offering paid tools, analyzing how to gate execution on successful payment without degrading user experience or LLM tool invocation reliability. We describe and compare seven coordination patterns—TWO_STEP, RESUBMIT, X402, ELICITATION, PROGRESS, DYNAMIC_TOOLS, and an emerging URL_ELICITATION variant—characterizing their assumptions, message flow mechanics, UX/security trade-offs, and common failure modes (e.g., tool misrouting, argument drift, hidden error payloads, client timeouts, and inconsistent rendering of progress messages). The work contributes: a capability-aligned design taxonomy for payment gating in MCP; a lightweight reference wrapper (PayMCP) that enforces “execute only after paid” without duplicating business logic and supports multiple gating modes (including AUTO selection based on client capabilities); a non-empirical compatibility snapshot synthesizing client-declared capabilities and observed initialization payloads, with practical notes on portability. The paper focuses on engineering design and coordination reliability rather than payment economics. It is intended for practitioners building paid MCP tools and for client/server developers evaluating portable, safe payment hand-offs across heterogeneous MCP clients.