Model Context Protocol unveils roadmap for agentic workloads and agent identity

The Model Context Protocol (MCP) has released an updated roadmap outlining the direction for protocol work over the coming months. Developed by Core Maintainers and community Working Groups, the plan identifies five priority areas: modern agentic workloads, unified transport mechanisms, agent identity and authorisation, tool calling improvements, and SDK development. The roadmap marks a shift from previous on-horizon items, elevating areas such as server-initiated events and agent identity to top-tier priorities as the protocol matures.

A central focus of the new plan is adapting to modern agentic workloads that no longer fit standard request-and-response patterns. The protocol aims to mature the Tasks extension, known as SEP-2663, for inclusion in the specification to support longer-running agentic loops. This work includes refining primitives such as subscriptions and progress notifications, ensuring that servers can push streamed results and clients can steer work mid-flight without relying on inefficient polling.

In transport mechanisms, the roadmap builds on the 2026-07-28 release, which standardised remote MCP servers as standard HTTP workloads. The protocol now seeks to unify this model to cover other deployment modes, including local servers speaking Streamable HTTP over stdio. This unification is intended to simplify server and client development, allowing MCP to operate seamlessly on infrastructure already used for APIs and services.

Addressing the growing complexity of cloud-based agents, the roadmap prioritises standardised agent identity and authorisation. Current MCP authorisation is designed for interactive clients where a person approves access in a browser, which is less suitable for cloud workloads acting on behalf of users. The new plan includes finalising Demonstrating Proof of Possession (DPoP) and defining paths for Workload Identity Federation, ID-JAG grants, and standard token exchange to replace pasted API keys and long-lived tokens.

To manage the scale of tool selection, the protocol is initiating a progressive discovery effort. This allows servers to reveal tool catalogs incrementally, reducing the computational cost associated with large tool lists that can degrade model performance. Additionally, the roadmap aims to standardise result handling in tool calling, providing a clear contract for how outputs are presented to models and addressing inconsistencies in current response formats.

The Model Context Protocol has engaged with standards bodies, including the IETF OAuth and WIMSE working groups, to help underlying standards evolve with the building blocks required for agent identity. Specification Enhancement Proposals (SEPs) falling within these five priority areas will receive expedited review, while those outside the roadmap will compete for scarce maintainer time. The protocol encourages developers to identify the relevant priority area for their proposals and engage with the corresponding Working Groups.