I found this excerpt from Did Skills Kill MCP? really helpful:
Saying skills killed MCP is about as accurate as saying GitHub Actions killed Bash…
MCP is where capability lives. It’s what allows an AI agent to actually do things instead of just talking about them. When an agent can run shell commands, edit files, call APIs, query databases, read from drives, store or retrieve memory, or pull live data, that’s MCP at work. MCP Servers are code. They run as services and expose callable tools. If an agent needs to interact with the real world in any meaningful way, MCP is almost certainly involved.
For example, if an agent needs to query the GitHub API, send a Slack message, or fetch production metrics, that requires real integrations, real permissions, and real execution. Instructions alone can’t do that.
Skills live at a different layer. Skills are about process and knowledge. They’re markdown files that encode how work should be done. They capture team conventions, workflows, and domain expertise. A Skill might describe how deployments should happen, how code reviews are handled, or how incidents are triaged. This is institutional knowledge made explicit.
- Angie Jones,
