Since our last November release MCP continued to grow at an astonishing rate. Across our Tier 1 SDKs, we’re seeing close to half-a-billion downloads a month, with both TypeScript and Python SDKs crossing the 1 billion total downloads threshold. In just a few months, the protocol continued to grow as the data and interactivity substrate for agentic workflows.
Today, we’re officially pushing the release button on the next version of the MCP specification, 2026-07-28, along with the SDKs that will allow you to start building clients and servers right away.
The highlight of this release is a stateless protocol core - MCP is transforming from a bidirectional stateful protocol into a request/response stateless protocol. It was one of the most highly-requested features from developers who were eager to get better reliability and scalability for their MCP servers.
A short demo of the stateless protocol core in action.
There is, of course, more to what we’re introducing with this version:
- Every request is self-describing, with an optional discovery call for clients that want capabilities up front, so any request can land on any instance behind a plain round-robin load balancer.
- Method and tool names travel in the
Mcp-MethodandMcp-NameHTTP headers, so gateways can route and authorize on headers directly. - Server-to-client requests for things like sampling and elicitation are being redesigned to use Multi Round-Trip Requests (MRTR), removing the need for constantly open bidirectional streams.
- List responses carry cache hints and a deterministic order, so clients can cache tool catalogs and keep upstream prompt caches stable across reconnects.
- Formally locking in on a proper extensions framework, with Tasks joining other extensions, such as MCP Apps and Enterprise Managed Authorization (EMA).
- A set of authorization hardening changes including RFC 9207 issuer validation and a formal shift away from Dynamic Client Registration (DCR) toward client metadata documents (CIMD).
- A formal deprecation policy with a twelve-month minimum window so you can plan upgrades instead of reacting to them.
The TypeScript, Python, Go, and C# SDKs are updated to match, with detailed migration notes for the breaking bits - and you can get started with the new spec right away.
What changed
No handshake or sessions
With the new spec version, we’ve officially retired the initialize/initialized exchange along with the Mcp-Session-Id header (refer to SEP-2575, SEP-2567). Each request now travels on its own, carrying its protocol version, client identity, and client capabilities in _meta. If a client wants to learn a server’s capabilities before doing anything else, there’s a new server/discover Remote Procedure Call (RPC) for that; however, it is not required. Any request can now land on any server instance behind a plain round-robin load balancer without needing shared storage.
POST /mcp HTTP/1.1
MCP-Protocol-Version: 2026-07-28
Mcp-Method: tools/call
Mcp-Name: search
{"jsonrpc":"2.0","id":1,"method":"tools/call",
"params":{"name":"search","arguments":{"q":"otters"},
"_meta":{"io.modelcontextprotocol/clientInfo":{"name":"my-app","version":"1.0"}}}}
Dropping the protocol-level session doesn’t force your application to be stateless. If your server needs to carry state across calls, mint an explicit handle from a tool and have the model pass it back as an argument. We found this works better than session state hidden in the transport - the model can see the handle and thread it between tools.
Multi Round-Trip Requests (MRTR)
MRTR replaces the server-initiated elicitation/create, sampling/createMessage, and roots/list requests that previously required a held-open stream.
Sometimes a tool needs something from the user mid-call, such as a confirmation or a missing parameter. MRTR (SEP-2322) enables this scenario over a stateless protocol: the server returns resultType: "input_required" along with the requests it needs answered, and the client retries the original call with the answers attached in inputResponses.
Streamable HTTP requests now must include Mcp-Method and Mcp-Name (SEP-2243). Your gateway, rate limiter, or WAF can route and meter on those headers instead of parsing JSON bodies.
List results are cacheable
Responses from tools/list, prompts/list, resources/list, and resources/read now carry ttlMs and cacheScope (SEP-2549). This allows clients to determine the best caching strategy for responses and reduce unnecessary re-fetching.
Authorization
From our discussions with implementers for the past year, authorization is where implementers spend most of their integration time. With this spec revision, we continued evolving the MCP auth and security posture.
- Authorization servers should return the
issparameter per RFC 9207, and clients must validate it before redeeming a code (SEP-2468). This closes an authorization-server mix-up hole. - Clients set
application_typeduring Dynamic Client Registration (DCR) so authorization servers stop rejectinglocalhostredirects for desktop and CLI apps (SEP-837). If you’ve ever wondered why your CLI client’s OAuth flow got aredirect_urierror, this is likely why. And while we’re moving to Client ID Metadata Documents (CIMD) as the standard, this is a hardening measure making the protocol comply with OAuth spec requirements. - Client credentials are bound to the issuer that minted them. No reuse across authorization servers (SEP-2352).
- Dynamic Client Registration itself is now formally deprecated in favor of CIMD. DCR continues to work for backward compatibility, but will be removed in a future version of the MCP spec.
Tasks
Tasks move out of the experimental core and into the io.modelcontextprotocol/tasks extension, with a poll-based tasks/get and a new tasks/update (SEP-2663). Change notifications move from the old HTTP GET endpoint to a single subscriptions/listen stream that clients opt into per notification type.
Deprecations
Roots, Sampling, and Logging are deprecated (SEP-2577). They still work, and they’ll keep working for at least twelve months. New implementations shouldn’t adopt them. The legacy HTTP+SSE transport is also considered to be officially deprecated, with a year-long offramp.
SDKs
All four Tier 1 SDKs speak 2026-07-28 as of today:
Beyond the Tier 1 set, the Rust SDK supports the new spec in beta.
The SDKs implement APIs that allow you to build both servers and clients with the new spec version. As we mentioned in the SDK beta blog post, there will be some migration cost, especially for developers that did depend on session identifiers; however, we incorporated early testing feedback that makes this process much easier.
Ecosystem support
As with any large release, the work that we’re doing with MCP would not be possible without contributions from folks across the ecosystem. We’re also especially grateful to a number of contributors and partners who helped us test and validate the spec before it became generally available.
David Soria Parra Member of Technical Staff, Co-Inventor of MCP
Alex Salazar CEO & Co-Founder
Swami Sivasubramanian VP of Agentic AI
Brendan Irvine-Broque Senior Director Product Management
Josh Clemm VP of Engineering
Anna Berenberg Engineering Fellow
Austin Parker Director of AI Strategy
Enrico Toniato CTO
Tina Schuchman Corporate Vice President for Engineering, Microsoft Foundry
Sean Roberts VP of Applied AI
Nick Cooper MTS & MCP Core Maintainer
Paul D'Ambra Product Engineer
Jeremiah Lowin CEO
Inian Parameshwaran Head of Product
Andrew Goodman VP of AI
Getting started
We’re excited to have developers build on the new spec. To get started, refer to the following resources:
Thank you
This release would not be possible without a massive community of contributors and industry partners. We’d like to acknowledge the dozens of key contributors across specification, documentation, SDKs, working and interest groups, as well as hundreds of independent worldwide communities who rallied support and excitement for MCP. We’re looking forward to continuing evolving the protocol as it grows!