Updated September 10, 2026. Google Cloud introduced a developer plugin for AI coding agents that packages agent skills and tools, including MCP-based capabilities, into an installable bundle. Google frames plugins as a way to manage capabilities that otherwise become fragmented across individual skills, documentation and remote MCP servers.
This suggests a useful separation of concerns: a model supplies general reasoning, skills provide task-specific procedure and context, MCP exposes executable tools or remote resources, and a plugin becomes the distribution unit that installs a coherent set of those capabilities.
A practical review checklist
- Inventory which instructions belong in skills and which operations require tools.
- Keep tool permissions independent from descriptive skill content.
- Version plugin bundles so teams can reproduce agent behavior.
- Avoid loading every skill into context when only a subset is needed.
- Treat third-party plugins as code and permission-bearing dependencies.
Sources and verification
Primary source: Google Cloud Blog — “Introducing the Google Cloud Developer Plugin for AI Coding Agents” (September 10, 2026).
This article distinguishes confirmed release facts from engineering interpretation. Dynamic details such as support status, limits and availability should be rechecked against the primary documentation before a production change.
