meal-planner-mcp: Command for Claude Code

You are considering adding a new MCP tool to this server: $ARGUMENTS

Before writing any code, apply this repo's four-part test (see CLAUDE.md → "Why this is an MCP and not just a prompt"). A capability earns a tool ONLY if it does something the model can't do well on its own:

1. Persists state across sessions (e.g. cooked-history in state.json)
2. Causes a side effect / writes a durable artifact (e.g. an export file)
3. Accesses private or live data (e.g. the user's local recipe library)
4. Computes something that must be exact (e.g. ingredient merge/scale math)

Then do exactly this:

• State, in ONE sentence, which of the four (if any) "$ARGUMENTS" serves.
• If it serves none — it's "reason over what Claude already knows" — do NOT add a tool. Show how to get the result by composing existing tools (e.g. suggest_recipes → swap_meal for "make Friday quicker"), and stop.
• If it serves at least one, scaffold it the repo's way:
  • Put the real logic in the appropriate PURE module (ingredients / core / planner / exports) with a direct unit test in tests/ — no I/O in the logic.
  • Add a THIN @mcp.tool() wrapper in server.py that calls the pure function and touches the filesystem only through store.py.
  • Give it a planning-flavored name that won't collide with a storage-style recipes server (see CLAUDE.md → "Running alongside another recipe MCP").
  • Mirror the wrapper into cli.py if it makes sense as a CLI verb too.
  • Run python -m pytest -q and ruff check src/mealplanner tests before declaring done.

Report which branch you took (composed vs. added) and the one-sentence justification.