The Model Context Protocol project has published an updated roadmap focused on long-running agent work, broader deployment options, agent identity, clearer tool results and stronger software development kits. Core maintainers developed the plan with community maintainers and working groups for specification releases expected over the coming months.

The roadmap follows the MCP specification issued on July 28, 2026. That release removed protocol-level sessions and the initialization handshake, allowing remote servers to scale without maintaining connection state. It added `server/discover` so clients can identify supported versions and capabilities before normal interaction, and made list results cacheable.

Agent communication also changed. Tasks moved into an official extension after feedback from early adopters, while multi-round-trip requests replaced server-initiated requests for flows such as elicitation on stateless servers. Governance work introduced a contributor ladder, delegated proposal triage to working groups and established feature lifecycle and deprecation policies. Authorization changes included issuer validation, issuer-bound client credentials and Client ID Metadata Documents as the preferred client-registration route.

The new plan groups future work into five areas. The first addresses agentic workloads that run longer than a conventional request and response. Maintainers want server-initiated events through mechanisms such as webhooks and channels, a review of how existing agent and transport features compose, and further development of the Tasks extension.

A second area seeks to extend the stateless HTTP approach to other deployments, including local servers using Streamable HTTP over standard input and output. The goal is a more unified transport model for clients and servers.

Identity is another priority. Current authorization commonly assumes a person approving access in a browser, while cloud agents may operate without a user present or delegate limited authority to sub-agents. Planned work includes proof-of-possession, workload identity federation, the ID-JAG grant associated with enterprise-managed authorization and standard token exchange, using existing identity standards where possible.

For tools, maintainers want one clearer output contract and progressive discovery. A server with hundreds of tools can otherwise expose its entire catalog before a user has stated a need, increasing context use and potentially making selection harder. Progressive discovery would initially reveal a smaller surface and expand it as a conversation narrows.

The fifth area covers SDK ergonomics, documentation and specification conformance across supported languages. Together, the priorities move MCP beyond basic tool calls toward infrastructure for persistent, discoverable and delegated agent activity, while the roadmap remains a statement of direction rather than a guarantee that every item will land in the next specification.

Maintainers assigned core owners and working groups to each area. That structure connects the technical agenda to the project's governance changes, giving individual proposals a route through review, lifecycle and eventual deprecation processes.