01 / DECIDE
Architecture decision
Use MCP when data must be live; use a Skill when the method should travel with the team and run against local context.
MCP connects authenticated external tools and context; a Skill packages instructions, scripts, references, and a trigger for a reusable workflow. Map input, output, state, and side effects as one observable path before deciding which layer owns choose mcp for live access and skills for repeatable method.
02 / BUILD
Three-step implementation
- 01
Frame the contract
Write down the caller, data classification, success condition, timeout, cancellation, and ownership. Use MCP when data must be live; use a Skill when the method should travel with the team and run against local context.
- 02
Build one narrow path
Implement one end-to-end path with request correlation, typed state, and reversible failure handling. MCP connects authenticated external tools and context; a Skill packages instructions, scripts, references, and a trigger for a reusable workflow.
- 03
Prove the outcome
Turn acceptance into a repeatable fixture, contract test, or browser test. Positive and negative trigger tests select the right Skill, while MCP diagnostics verify identity and failure behavior.
03 / BOUND
Production boundary
Minimize MCP tools and credentials; keep Skills auditable, scoped, and explicit about scripts and dependencies.
04 / PROVE
Acceptance evidence
Positive and negative trigger tests select the right Skill, while MCP diagnostics verify identity and failure behavior.
SOURCE / HTTP
Reproducible source probe
curl -fsSI 'https://learn.chatgpt.com/docs/extend/mcp' | sed -n '1,5p'