Skip to main content
5.8

The Mission

Estimated time: 40 minEstimated cost: ~$0.30Tool: Claude Code (Sonnet)
After this drill, you can:

After this drill, you have run a complete MCPP-lite cycle — Intent → Plan → Execute → Report — and understand the pattern behind all effective agentic engineering.

Why this matters

MCPP-lite is the distillation of everything you learned in Module 5 into five words: WHAT, CONTEXT, DON'T, GO, REPORT. It is the minimum viable agentic protocol — a format that works for Claude Code, for Claude Desktop, for future AI agents you have not met yet. This drill runs the complete cycle on a real task and ends with a reflection: you now have a pattern that scales to any problem you can describe well.

How to do it

  1. 1

    Choose a real, meaningful task for your project

    Something you genuinely want done — not a test. This is the task that will go into your capstone if it is good enough.

  2. 2

    Write the full MCPP-lite brief

    Use the five-section format: WHAT / CONTEXT / DON'T / GO / REPORT. Each section must be complete.

  3. 3

    Run the brief through Claude Code — full autonomous execution

    No mid-session intervention. If it asks a question, answer it and tell it to continue. This should be mostly hands-off.

  4. 4

    Review the report and fill out the retrospective

    What was in the report? What would you change in the WHAT section? What missing CONTEXT caused any issues? Update CLAUDE.md if you learned something about your project.

The prompt

PROMPT — Full MCPP-lite Mission BriefModel: Claude Code (Sonnet)Est. cost: ~$0.30
WHAT: [The desired outcome — specific, verifiable, described as a user behavior or system state]

CONTEXT: [What Claude Code needs beyond CLAUDE.md — relevant files, constraints, background, decision rationale]

DON'T: [Hard constraints — files not to touch, patterns not to break, dependencies not to add, things explicitly out of scope]

GO: Execute this autonomously. Make all implementation decisions consistent with the project conventions in CLAUDE.md. If you encounter an ambiguous situation, choose the option most consistent with the existing codebase patterns.

REPORT: When complete:
1. Summary: what was built in one sentence
2. Files changed: list with one-line description of each change
3. Decisions made: any non-obvious choices you made and why
4. Verification: how to confirm it works (what to click/test)
5. Follow-up: anything that should be done next or that needs review

Success criteria

  • ✓You ran the complete MCPP-lite format with all five sections
  • ✓Claude Code executed autonomously and produced a structured report
  • ✓You completed a retrospective identifying what you would change in the brief
  • ✓You updated CLAUDE.md if you learned anything new about your project

Common mistakes

Writing a CONTEXT that duplicates CLAUDE.md

→ CLAUDE.md is always read. CONTEXT is for session-specific additions: what is different about THIS task versus the general project description in CLAUDE.md.

Vague DON'T constraints

→ "Don't break anything" is not a constraint. "Do not modify src/lib/auth/ — that file is locked" is a constraint. Name the specific thing that must not change.