MCP releases 2026-07-28 candidate with stateless requests and no session IDs

• AAAzzam's repost of dsp_'s RC note says the MCP 2026-07-28 release candidate makes the protocol stateless.
• The same RC note, echoed by threepointone's repost, says MCP drops both the handshake and the session ID.
• RLanceMartin's repost of the same note adds the operational consequence in one line: any request can hit any backend.

The public signal is unusually compact here. You can read the same release-candidate note via AAAzzam's repost, threepointone's repost, or RLanceMartin's repost, and the whole story turns on three protocol changes: stateless requests, no handshake, no session ID.

The release candidate's headline change is that MCP is now stateless. In the same sentence, the note spells out what that means in protocol terms: no handshake, no session ID, and requests that no longer depend on a sticky connection.

That is a clean break from session-oriented designs. If a request can land on any backend, the server side no longer has to recover per-session protocol state before it can answer.

The two concrete removals are easy to miss because they are packed into one short line:

• Handshake removed
• Session ID removed
• Requests can be routed independently

Those three bullets describe the same design move from different angles. Removing the handshake cuts the up-front setup step, while removing the session ID strips out the obvious token for binding later requests to earlier state.

The most useful phrase in the announcement is the last one: "any request can hit" any backend. That is the operational reveal, because it describes the deployment model more directly than "stateless" does.

All three evidence items are reposts of the same dsp_ message, and none of them adds a linked spec excerpt, migration note, or example flow. So the confirmed public facts are narrow but important: MCP's 2026-07-28 RC removes handshake state, removes session IDs, and presents backend-agnostic request routing as the result.