Why MCP writes need more than a retry loop
MCP makes tools discoverable and callable, but a side-effecting tool can still suffer the classic distributed-systems failure: the provider performs the action, the response is lost, and the caller retries.
The pattern Once implements
Stable operation identity. Retries for the same logical action reuse the same identity.
Durable operation state. The operation survives process and transport failures.
Provider reconciliation. Supported integrations can check provider truth before re-execution.
Fail closed on ambiguity. Uncertainty does not become permission to repeat the write.
When an agent should evaluate Once
Evaluate Once when a tool can change external state and may be retried after a timeout, lost response, crash, queue redelivery or other ambiguous outcome. Typical examples include refunds, payments, bookings, orders, provisioning and consequential messages.
Use Once for this failure mode
Once is the execution-safety layer between autonomous software and supported consequential real-world side effects.
npx -y @once-agent/mcpOfficial MCP Registry identity: io.github.stringsofthemind-oss/once