A client supports only the current revision and is being pointed at servers of unknown vintage over stdio. The team debates whether to probe with discovery first or just send the request they want. What does the era model say?

The principle — When two eras of a protocol share a channel, the danger is not rejection but a counterpart that misunderstands you politely.

Why the key is correct — The era model names three kinds of implementation. Legacy: protocol versions that establish a session with an initialize handshake (2025-11-25 and earlier). Dual-era: an implementation that supports both modern and legacy versions. A client that supports only the current revision is neither of those, so it has no fallback of its own, and that raises the cost of guessing wrong. The reason to probe is that a legacy server does not reliably refuse an unexpected request: some will process a method that exists in both eras under legacy semantics, producing a result that looks like success. A probe converts that into a deterministic answer, and the answer is worth caching because the era determination is a property of the server, not of an individual request.

Why the others are wrong — Legacy servers do not answer under modern semantics, and clean rejection is not guaranteed. The era is a property of the server rather than of each request. And falling back to the handshake requires dual-era support this client does not have.

Remember this — Ambiguous success is worse than a clean failure, and probing is how you buy the clean failure. Cache the answer for the lifetime of that server rather than deciding again on every request, and re-probe only if the cached assumption later stops holding.

Sources — MCP first-party documentation.