In a recent keynote by David Soria Parra from Anthropic, insights were shared about the transformative potential of the Message Control Protocol (MCP) and how it will shape the development of AI agents by 2026. As we delve into the connectivity stacks and skills needed for future agents, a vision for an interconnected, versatile AI ecosystem begins to unfold.
Introduction
Based on content from AI Engineer
David Soria Parra opens with a powerful statement: the future of AI agent interface won’t rely on plugins or hardcoded systems. Instead, we’re heading towards a model served over an MCP server. This change allows agents to deploy seamlessly across platforms like ChatGPT and VS Code Cursor, enhancing functionality through MCP’s intrinsic protocol-driven environment.
The Evolution of MCP
Reflecting on the past 18 months, the MCP ecosystem has witnessed unprecedented growth, moving from a spec document to a thriving framework with remote capabilities and a central authorization system. Thanks to contributions from major players like OpenAI and Google, MCP’s reach now includes over 110 million monthly downloads.
From Exploration to Production
The transition from experimentation in 2025 to production in 2026 marks a pivotal shift. In the coming years, agents will evolve from localized entities focusing on coding to powerful tools capable of handling varied tasks across domains like finance and marketing. The emphasis will be on connectivity, with the right tool for the right task being paramount.
The 2026 Connectivity Stack
The foundation for building agents lies in a robust connectivity stack comprising skills, MCP, and CLI/computer use. Each component serves specialized functions, though MCP stands out by offering platform independence and rich semantics vital for enterprise-level integration.
Progressive Discovery and Improved Client Harnesses
To maximize these capabilities, Parra emphasizes the importance of progressive discovery. By leveraging this pattern, developers can optimize context load, deferring the loading of tools until necessary, thereby improving efficiency and reducing resource consumption.
Programmatic Tool Calling and Agent Orchestration
Programmatic tool calling marks another leap forward, advocating for composed executions to reduce latency and enhance functionality. By incorporating mechanisms like structured output and repple tools, developers can streamline operations and achieve seamless integration.
Roadmap for MCP and Strategic Integrations
The upcoming MCP extensions and enterprise features aim to refine asynchronous tasks and improve core scaling capabilities. Collaborative efforts with hyperscalers and innovations like the stateless transport protocol underscore the strategic future of MCP. Furthermore, new cross-app access features are under development to facilitate enterprise adoption.
Conclusion and Community Feedback
As the keynote concludes, the call for community engagement is clear. As we build towards a future where agents effortlessly navigate complex connectivity landscapes, feedback and collaboration will play a crucial role in refining these technologies and realizing the full potential of AI innovation.
