The Model Context Protocol (MCP) is rapidly becoming the backbone of AI agent integration, and the recent MCP Dev Summit in New York made one thing clear: the race to standardize AI agents is officially underway. With over 110 million SDK downloads per month, MCP has achieved in 16 months what it took React three years to accomplish—and now the industry is grappling with what comes next.
The Numbers Tell the Story
The Agentic AI Foundation (AAIF), the Linux Foundation sub-foundation overseeing MCP standardization, announced impressive growth metrics: 170 member organizations, 1,200 attendees at the summit, and seven active working groups with more than 500 participants. The Identity and Trust Working Group alone has 150 members from 66 organizations—a sign that enterprise concerns are front and center.
“When you move from experimentation to production, and you have a thousand people doing their own thing and showing you some cool demos and moving to production at scale, you do need an open standard. No one company can define it, and no one company can own it,” said AAIF Executive Director Mazen Gilbert.
Security Remains the Top Enterprise Concern
Despite rapid adoption, organizations are clear about what they need before deploying MCP in production: reliability, scalability, governance, and enhanced security. MCP maintainers acknowledged these requirements and announced the AAIF is soliciting proposals to sponsor working groups addressing enterprise needs.
“We’re happy to have AAIF bring the industry together and talk about the right solutions across the agentic ecosystem,” said David Soria Parra, Technical Staff at Anthropic and co-creator of MCP.
The protocol roadmap includes several key enhancements: stateless transport to facilitate hyperscaler-friendly deployments, task primitives for asynchronous and long-running agent operations, enterprise authentication with cross-org token exchange, server-initiated triggers, and skills for bundling domain-specific knowledge.
Beyond MCP: The Broader Agentic Ecosystem
Nick Cooper, Technical Staff at OpenAI and MCP maintainer, emphasized that MCP is just the beginning: “MCP is the seed. The foundation has a broad mandate beyond just MCP.”
The working groups are already tackling broader challenges like context bloat (using progressive discovery and tool search for deferred loading) and composability (structured outputs that let models chain tool calls without round-tripping to inference).
Looking Ahead
With expanded AgentCon and MCP Con events planned for Europe and North America in 2026, the standardization of AI agents is accelerating. The question is whether the AAIF can move fast enough to prevent industry fragmentation before the ecosystem becomes too fragmented to unify.