If you read one line: The MCP spec finalizes Tuesday (7/28), but the founder's story isn't the date — it's that the ecosystem to catch you already shipped. Four production SDKs, a live registry, and zero-touch enterprise auth are out now, all verifiable from primary sources. Nothing you run breaks Tuesday. Your weekend is three moves: test a beta client, publish to the registry, wire enterprise auth.
Most of this week's MCP coverage — ours included — points at one calendar square: July 28, when the stateless spec goes final. That's the right thing to watch. It is not the thing to do. The deadline is a governance event; the work a founder can act on this weekend is already sitting in package registries and a live API. Here's the verified picture, source by source.
1. The spec: a deadline that asks nothing of you today#
The 2026-07-28 revision is a published release candidate right now, scheduled to finalize Tuesday (MCP blog). It's the largest rewrite in the protocol's history — stateless transport, an Extensions framework, hardened OAuth, and a set of deprecations.
What it means for you: the deprecation policy is the part that lets you exhale. Removed features keep working for at least a ~12-month window, and a new client falls back to the old handshake when it meets an old server. So a server you deployed months ago does not break Tuesday morning. We walked the exact server-author changes in the day-one migration checklist and argued the real headline is governance, not statelessness. Read those for the how. This piece is about the what's-already-here.
2. Four SDKs are out — in four languages — before the date#
You don't have to hand-roll the new protocol. Beta SDKs shipped ahead of finalization in Python (v2, with an overhauled MCPServer API), TypeScript (v2, now split into @modelcontextprotocol/server and @modelcontextprotocol/client), Go (v1.7.0-pre.1), and C# (v2.0.0-preview.1) (MCP blog). Existing servers keep running through the fallback, so upgrading is opt-in.
What it means for you: you can validate against the future without leaving the present. Keep production on the stable line — the current stable Python package is mcp v1.28.1, released June 26 (PyPI) — and point a beta client at your server this weekend to see what actually moves. We broke down what installs, breaks, and migrates in the betas. The move is a dry run, not a rewrite.
3. There's a real front door now: the official registry#
Discovery stopped being a README problem. The official MCP registry is live, and its /v0/servers API returns active, timestamped public listings with cursor pagination (registry API).
What it means for you: this is distribution you don't have to earn one retweet at a time. If you've built a server, publishing it to the registry puts it in the channel clients actually query — and it's a different decision from distributing the runtime via OCI, which is about shipping the bits, not being found. For a solo builder, "listed where the agents look" beats "starred on GitHub."
4. Enterprise auth got boring — which is the point#
Enterprise-Managed Authorization reached stable: zero-touch OAuth where a customer's IT admin grants your server access through their identity provider — ID-JAG tokens minted from SSO, Okta the first supported IdP, with Claude, VS Code, and servers like Asana, Figma, and Linear already on board (MCP blog).
What it means for you: the per-user consent screen that quietly kills enterprise pilots is now optional. A buyer's admin flips one switch and every seat is provisioned. Supporting EMA via ID-JAG is the cheapest "enterprise-ready" badge a one-person company can earn — no SOC 2 auditor required to start.
The spec is the headline. The SDKs, the registry, and the auth are the business. A deadline you can't act on is news; a distribution channel you can publish to today is leverage.
5. The through-line: MCP stopped being a spec and became a platform#
Zoom out and the week's real signal is maturity. A protocol with four production SDKs, a live registry, an enterprise auth story, and a 12-month change-notice guarantee is no longer a moving target you hesitate to build on — it's infrastructure. That's the same shift that makes the boring stuff (auth, discovery, versioning) the interesting stuff. If you'd been deferring an MCP integration waiting for the ground to stop moving: Tuesday is the ground stopping.
And it frees your attention for the layer above the protocol — keeping a long-running agent inside its context window, still the most-read engineering problem we cover. The plumbing is settling so you can go back to the part that's actually hard.
Your weekend in three moves#
- Dry-run a beta client. Point a v2 SDK client at your existing server and watch what falls back vs. breaks. Cost: an afternoon. Payoff: no surprises when you migrate on your own clock.
- Publish to the registry. If you have a server worth using, list it at
registry.modelcontextprotocol.io. Distribution you don't have to beg for. - Wire EMA if you sell to companies. ID-JAG over the buyer's SSO turns a stalled pilot into a one-admin-click rollout.
The deadline is Tuesday. The leverage has been on the pad for weeks. Go use it.
