nutrition

name	nutrition
description	Look up nutrition data for foods, plan meals, calculate macros. Use when discussing calories, protein, fat, carbs, dietary analysis, food comparisons, or barcode scanning.
argument-hint	["food name","meal description","or barcode"]
user_invocable	true

/nutrition — Food Nutrition Lookup

Analyze nutritional content of foods using the nutrition-mcp tools.

When the user invokes this skill

1. Determine query type and run the lookup in a single step:

   • All digits (12-13 chars) → nutrition_barcode
   • Otherwise → nutrition_search

2. Check the results for seeding status: if the response includes a seeding progress message (e.g. "Local database is being downloaded" or "X% imported"), the local database is still being populated. Show the progress to the user alongside any USDA results. If seeding hasn't started yet (no progress message and no local-tier results), call nutrition_seed to start it, then inform the user that the local database is being set up in the background.

3. Check for USDA key: if source_tier is only "local" for all results (no "usda" tier), append a note:

   Note: USDA API key is not configured — results are local-only. Get a free key at https://fdc.nal.usda.gov/api-key-signup and add USDA_API_KEY=your_key to .env for broader coverage.

4. Present results in a table showing name, brand, and per-100g macros (calories, protein, fat, carbs). Include fiber/sugar/sodium when available. Note the source (local, USDA, or cached).

5. If the user wants details on a specific result, look up the full record by ID.

6. If results are missing, incomplete, or a poor match (e.g. no results, only generic matches for a specific branded product, or the food is region-specific / niche):

   • Call nutrition_cache_stats to check seed and USDA key status.
   • If seed.phase is "idle" and local foods count is 0: call nutrition_seed to start seeding, then tell the user the database is being populated and to try again in a few minutes.
   • If seeding is in progress: show the progress percentage and suggest waiting.
   • If seeding is done and local data exists: search the web for the food's official nutrition information (manufacturer's website, product page, or a reputable nutrition database). Then cache the verified data using nutrition_cache_add with the source URL. This ensures the user gets accurate, up-to-date nutrition data rather than estimates.

7. If the user says a result looks wrong (e.g. "that can't be right", "those calories seem off", "this is the wrong product"):

   • Web search for the food's official nutrition information from a credible source (manufacturer label, product page, government database).
   • If you find corrected data: use nutrition_override with the corrected fields and the original food's ID. This preserves the original entry and creates a corrected copy. Tell the user what you changed and cite the source.
   • If the cached entry is simply the wrong product (not a data error): delete it with nutrition_cache_delete, then search the web for the correct product and cache it with nutrition_cache_add.
   • If you cannot find better data, tell the user and suggest they provide the correct values manually.

8. Browsing cached entries: If the user wants to review what's been cached (e.g. "show me my cached foods", "what have I saved"), use nutrition_cache_list. Filter by tier ("usda", "web", or "all") as appropriate.

9. IMPORTANT — never guess nutrition values from training data. Nutrition information must come from one of these sources:

   • The local database or USDA API (via the nutrition tools)
   • A web search of a credible source (manufacturer label, government database, registered dietitian resource)

   If you cannot find reliable data from any of these sources, tell the user what you searched and that you couldn't find verified nutrition information. Do not fill in values from memory — they may be outdated or inaccurate.

Examples

/nutrition chicken breast
/nutrition 012345678901
/nutrition compare salmon vs tilapia

Notes

• All values are per 100g unless a serving size is specified
• The local database must be seeded on first use via nutrition_seed (~3 min, runs in background)
• When a food isn't in the local DB or USDA, always web search for accurate data and cache it with nutrition_cache_add — do not rely on training data for nutrition values