- Generate SKILL.md with all the tools and their descriptions for ease of use by the agent.
	Make your MCP server “skill-friendly” with a few small technical tweaks
	•	Ensure tool names and schemas are stable and descriptive, because Skills rely on predictable interfaces.
	•	Provide a single “diagnostics snapshot” tool (even if it internally calls others) so Skills can start with one cheap probe.
	•	Include golden examples in your docs/prompts that Skills can copy, and keep them updated as your toolset evolves. This aligns with how Skills are described across Claude/Cursor/Codex: instructions + resources that steer behavior reliably.  

- A major trend in 2026 is giving agents tools to "think" before they act.
Sequential Thinking: Add a tool called `plan_step` that allows the agent to output its reasoning process into a structured format before calling a "real" tool. This improves reliability for complex DevOps tasks.
https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking
​

- kube_logs (multiple log streams + events, selectors support)
- add support for k8s context selector