MCP's new roadmap makes agent identity and enterprise auth the top priority

The Model Context Protocol, or MCP, is moving closer to the concerns that matter inside a real business: who an AI agent is, what it can access, and how those permissions are managed across company systems.

Its steering group has set five priorities for the next phase: server-pushed agent messaging, a unified HTTP transport, agent identity and enterprise security, progressive tool discovery, and better SDKs for developers. The security work is the key signal. It includes DPoP and Workload Identity Federation, with a move away from browser-based authentication. In plain English, MCP is being shaped for AI agents that need to operate under controlled, auditable access rules, not just as useful experiments on a developer’s laptop. You can see the roadmap discussion in the original report.

For operators, this matters because the value of AI agents rises sharply once they can work across internal tools, customer data, and business workflows. So does the risk. An agent that can read a CRM, query financial data, or trigger an operational process needs a clear identity and tightly defined permissions. Without that, scaling agent use across teams becomes hard to govern, particularly in regulated or security-conscious firms.

This roadmap doesn’t mean every company should deploy MCP tomorrow. It does mean the infrastructure underneath enterprise AI is starting to focus on the right problems. This is the kind of thing we build into an AI command centre, where access, oversight, and useful business workflows need to sit together.