From CLI to MCP in 20 Minutes
summary
Shows an automated path from the kubectl-mtv CLI to a working MCP server that reuses CLI parsing and help text to expose commands like get-plans over MCP, uses an inference server plus a base (granite) and instruction-tuned model to generate tool calls, and runs locally (port 8000) to browse, plan and run VM migrations via kubectl-mtv MCP browsing into OpenShift Lightspeed. Demoed server returns the same JSON error and CLI output, proving the generation approach works and arguing MCPs constrain models safely versus loose skills.
AI-written summary · about
chapters (12)
- 0:00Introduction to MCP
- 1:51Language model limits
- 3:19Why MCP is needed
- 7:09Base to assistant models
- 10:19Tool usage in LLMs
- 13:36Inference servers and APIs
- 17:30Tools to MCP protocol
- 19:20Why developers need MCP
- 21:32MCP vs CLI vs skills
- 25:24Live MCP server demo
- 27:56MCP vs CLI and skills
- 30:52Q&A
related talks
17mMCP Is Not Good Yet
10m⚡Lightning Talk: How I Built My Laptop Into an MCP Server To Create Secure Cloud Native Infrastructure
44mSupercharging DevOps with MCP (Without Opening a Security Hole)
26mBuild your First MCP App
32m30 minutes to understand MCP (Model Context Protocol)
29mBuilding MCP Apps With GraphQL Patterns You Already Know
28mBuilding MCP servers with VS Code — Level up your MCP
30mLet's make an MCP Server!