Is MCP dead? 2025 vs 2026 at GCP Next
Comparing the Google Cloud Next session catalogs — 2025 vs 2026: MCP: 1 → 63 sessions, A2A / agent-to-agent: 0 → 16, “Protocol”: 0 → 50. Read on for more details.
This is Google moving from “AI assistant” framing toward something much closer to an agent execution stack. The 63 MCP sessions span 32 Breakouts, 11 Lightning Talks, 8 Discussion Groups, 5 Demos, 3 Workshops, 3 Solution Talks, 3 Birds of a Feather, 1 Spotlight, plus a Developer Meetup and a Lounge Session.
The themes are pretty consistent:
• Gemini CLI as the developer control surface
• MCP as the protocol plumbing
• ADK as the agent toolkit
• BigQuery, Apigee, Cloud Run, and SecOps as the production surfaces
There’s also a quieter shift sitting next to MCP: the word “skills”: 26 sessions mention it in 2026. But the term is still in transition. Most of those sessions still mean “human training”, “certifications”, “upskilling programs”. But a smaller subset clearly means something else:
• Automating the UI with Gemini CLI, MCP and Skills
• Vibing with Gemini CLI: Custom Web AI solutions and shareable skills
• One tool to rule them all: Extending and customizing the Gemini CLI
• 10x productivity with the Gemini CLI
In those sessions, skills starts to mean reusable, executable capability — something you compose and ship, not just something a human learns in a course. The word hasn’t fully crossed over yet. But the direction is visible.
There’s also a demand signal in the reservations data: 7,778 seats already taken across the MCP sessions, weeks before the conference opens! The six biggest registration totals alone account for well over 1,400:
• Conversational Analytics agents, API, and MCP — 356 (Richard Kuzma)
• Agent development with BigQuery, ADK, and MCP — 298 (guillaume blaquiere)
• Build AI agents on Cloud Run — 256 (Ryan Pei)
• Google MCP Services — 178 (Romin Irani)
• Beyond the prompt: Build production-ready agents with Google’s MCP servers — 157 (Tarun Gumar)
• Unlocking enterprise actions: Bridge your APIs to AI agents with MCP and Apigee — 152 (David Rush)
And many of the smaller practitioner sessions are already gone (Agent development and AgentOps with BigQuery, ADK, and MCP; Building custom agents with Looker MCP and APIs; Build an AI Agent and Configure an MCP Server on Cloud Run)
Conference agendas are lagging indicators. CFPs, editorial planning, and speaker lock-in happen months before the event. So this doesn’t prove MCP wins long term. But it does show that well before the discourse settled, Google had already decided MCP was worth putting across one of its biggest stages.
And if MCP is the obvious signal, skills as executable capability may be the next term worth tracking.
→ Read more insights → Discuss on LinkedIn
