MCP 2026-07-28 Ships: Apps, Tasks & Governance, Revisited | TrueFoundry
Meet TrueForge: The open-source, vendor-neutral agent harness. 50% lower cost. Explore Now→
MCP 2026-07-28 Ships: Revisiting Apps, Tasks, and Gateway Governance Under the Largest Protocol Revision Since Launch
Published: August 6, 2026
Built for Speed: ~10ms Latency, Even Under Load
Blazingly fast way to build, track and deploy your models!
- Handles 350+ RPS on just 1 vCPU — no tuning needed
- Production-ready with full enterprise support
In late July we published our analysis of governing MCP Apps and Tasks at the gateway — and the protocol promptly gave the post a sequel deadline. On July 28, 2026, the Model Context Protocol’s final 2026-07-28 specification shipped on schedule, after the maintainers’ release-candidate description. The core goes stateless: protocol sessions and the Mcp-Session-Id header are gone, the initialization handshake is dropped, requests route on Mcp-Method and Mcp-Name headers, and applications that need state across calls can mint explicit handles — a basket_id, a ticket_id.
Key Takeaways
- The final MCP 2026-07-28 specification shipped on schedule as the protocol's largest revision since launch: a stateless core, a formal extensions framework, and a formal deprecation lifecycle.
- The two features our July analysis covered are now formal extensions — and Tasks' redesign is an explicit breaking change for anyone who built against the 2025-11-25 experimental API.
- Statelessness relocates state: applications that need continuity mint explicit handles that are visible in transcripts.
- Tasks formalize the temporal-validity problem: a durable task can outlive the authorization that started it.
- Apps put untrusted, server-rendered UI inside trusted client chrome — the security analyses flag user-targeted phishing through trusted AI interfaces as the new abuse avenue.
1. Before and After: What Actually Changed
MCP Specification Comparison
| Area | 2025-11-25 | 2026-07-28 final | Governance nuance |
|---|---|---|---|
| Core / transport | Stateful: sessions, sticky infrastructure | Stateless: header routing; plain load balancers | Session-ID hijacking removed; state visible as handles |
| Tasks | Experimental core feature | Formal extension: poll-based tasks | Durable work outliving its authorization |
| Apps | Emerging UI extension | Formal extension: server-rendered UI | Untrusted content in trusted chrome |
| Authorization | Dynamic Client Registration | Hardened toward deployed OAuth/OIDC practice | Client-side hardening is best implemented once |
| Notifications | Legacy HTTP GET change stream | Single subscriptions/listen stream | One stream to observe — and one to govern |
| Lifecycle | Informal evolution | Formal deprecation policy | A dated migration clock — inventory now or meet removals as incidents |
2. Three Nuances We'd Add to Our July Analysis
Where applications need continuity, explicit handles replace hidden transport state. A handle in model context is a capability-like artifact. Tasks turn temporal validity from a paper problem into designed behavior. Tasks make ongoing validity expected and affiliated with a specified principal.
Apps are a phishing surface with a saving grace. The architectural design of apps ensures that proxy calls remain governed, maintaining a layer of security.
3. The Migration Quarter: Where the Gateway Earns Its Keep
When the protocol itself is migrating, the layer every routed MCP call crosses becomes the inventory. Modern MCP requests carry protocol version, client identity, and capabilities metadata on every call. The gateway can implement client-side hardening solutions centrally, ensuring uniform compliance.
4. Boundaries, Stated Plainly
Sources and limits. The specification changes are documented in the official MCP release posts and the deprecated-features registry. The protocol just made its bid to be infrastructure — stateless, extensible, with enterprise-aligned auth.