What Happened
In late April 2026, as “MCP is dead” rhetoric grew louder, an Anthropic Staff Engineer publicly redefined MCP’s strategic position. The core thesis is clear:
MCP isn’t dead. MCP + Skills combined is the infrastructure that 2026 agents will run on.
The article garnered 42,750 views and 416 bookmarks on X, becoming one of the most discussed pieces in AI infrastructure that week.
Core Message Breakdown
1. MCP’s Positioning Shift
MCP (Model Context Protocol) is no longer viewed as an independent “tool connection standard” but as the foundational layer of agent runtime when deeply integrated with the Skills framework:
- MCP standardizes communication protocols between agents and external systems
- Skills defines agent behavior patterns and workflows
- Together, they form the agent’s “operating system”
2. From Coding Agents to Knowledge Worker Agents
The article clearly outlines the 2025 → 2026 paradigm shift:
| Dimension | 2025 | 2026 |
|---|---|---|
| Agent Type | Coding agents (Claude Code, Codex) | Knowledge worker agents |
| Coverage | Single development environment | Across 5+ SaaS applications |
| Work Mode | Write code, modify code | Cross-system coordination, decision-making, execution |
| Core Capability | Code understanding and generation | Multi-system tool invocation + context maintenance |
3. Google’s Follow-up
Almost simultaneously, Google Cloud announced 50+ fully managed remote MCP servers covering infrastructure, AI, databases, operations, security, documentation, Workspace, and full-stack services. This confirms MCP has become an industry-standard agent connection protocol, not just Anthropic’s proprietary protocol.
Why It Matters
An agent’s value isn’t in “chatting” — it’s in “getting work done.” And getting work done requires reliable connections to external systems.
Lessons from the past year:
- No matter how powerful a single LLM is, without a toolchain it’s just an ivory tower
- Fragmented tool protocols erode agent efficiency gains through integration costs
- MCP provides a standardized “USB-C port” for agents to plug-and-play with services
When MCP combines with Skills, agents go beyond simple “call an API” behavior to:
- Understanding task goals through Skill definitions
- Discovering available tools via MCP
- Orchestrating tool invocation sequences
- Maintaining context consistency across tool calls
Action Recommendations
- Immediately: Evaluate whether existing agent projects use MCP; if not, create a migration plan
- Short-term (1 month): Build MCP + Skills prototype, test cross-SaaS scenarios
- Medium-term (3 months): Build knowledge worker agents on MCP, covering 3-5 core business systems
- Ongoing: Monitor Google Cloud MCP Server ecosystem and Anthropic official Skill template updates
MCP isn’t the endpoint of tool protocols, but it’s currently the most promising path to becoming the industry standard.