rjvrai01-beep/calorie-counter

Calorie Counter Skill — FatSecret MCP v5

Purpose

Track daily food for Rajiv and Jasleen using LIVE FatSecret data via https://rajivfood.duckdns.org/mcp. All nutrition values, confidence scores and verification links come from this MCP server — never from Claude's own knowledge.

---

ABSOLUTE RULES — NEVER BREAK

1. ALWAYS call FatSecret MCP tools first — never use Claude's own food data
2. TABLE_ROW values from the server must be copied EXACTLY — never modify them
3. Every table row must have a Confidence column — values from TABLE_ROW only
4. Every table row must have a FatSecret Link column — values from TABLE_ROW only
5. If FatSecret returns no result — say "not found in FatSecret" — never estimate
6. NEVER say "the skill doesn't support X" — if a tool exists for it, call the tool
7. Delete = call FatSecret:delete_food_entry — never just redraw table without it
8. This skill requires Claude Sonnet or Opus — Haiku cannot reliably follow these steps

---

Daily Calorie Targets

| Person | Weekday (Mon–Fri) | Weekend (Sat–Sun) | |---------|-------------------|-------------------| | Rajiv | 1,650 kcal | 2,100 kcal | | Jasleen | 1,650 kcal | 2,100 kcal |

---

Step 0 — Identify Person

• If "R", "Rajiv", "J", or "Jasleen" found → proceed to Step 1
• If NOT found → ask "For Rajiv or Jasleen? (R/J)"
  • Remember the full original food command
  • When they reply R or J → continue with their original command automatically
  • Do NOT ask them to retype food

---

Step 1 — Parse Input

Extract from user message:

• Person: R/Rajiv or J/Jasleen
• Meal: breakfast / lunch / dinner / other (infer from time if not stated)
• Foods + quantities in ANY format:
  • add R 100 gm rice → rice, 100g
  • add R rice 100g → rice, 100g
  • add R 2 rotis and dal → roti × 2, dal × 1
  • add J 1 cup poha → poha, 1 cup
  • add R chicken biryani 250gm → chicken biryani, 250g
  • add J poha → poha, default serving

Build:

• foods list: individual food names → e.g. ["rice"] or ["roti", "dal"]
• qty_map: quantity per food → e.g. {"rice": "100g"} or {"roti": "2 pieces"}

---

Step 2 — Call FatSecret MCP (MANDATORY — NEVER SKIP)

Call: FatSecret:search_multi_food
Input: {
  foods:   ["rice"],           ← individual food names only
  qty_map: {"rice": "100g"},   ← quantities mapped per food
  region:  "IN"
}

The server returns structured blocks including:

• TABLE_ROW: — pre-formatted markdown table row (copy this DIRECTLY)
• CONFIDENCE: — score/100 + emoji
• FATSECRET_URL: — verification link
• STATUS: — AUTO_LOG, CONFIRM, or MANUAL_PICK

Act on STATUS:

• AUTO_LOG → copy TABLE_ROW into calorie table, then call log_food
• CONFIRM → show user: "I found '[food name]' for '[query]' — correct? (y/n)" If yes → copy TABLE_ROW_IF_CONFIRMED, call log_food
• MANUAL_PICK → show ALT options by number, wait for user choice

Then call log_food for each confirmed item:

Call: FatSecret:get_food { food_id: "[id from search result]" }
→ gets serving_id and exact per-serving nutrition

Call: FatSecret:log_food {
  food_id:    "[id]",
  serving_id: "[default serving_id from get_food]",
  quantity:   [number],
  meal:       "[meal]",
  date:       "[YYYY-MM-DD]"
}

Finally always call:

Call: FatSecret:get_day_summary {
  date:           "[YYYY-MM-DD]",
  calorie_target: 1650
}

Use the TOTALS_ROW and PCT_ROW from this response for the table footer.

---

Step 3 — Build the Calorie Table

Header (bold): Rajiv DD-MM Calorie Counter or Jasleen DD-MM Calorie Counter

Exact column order — always 8 columns (# is row number prefix):

| # | Food Name | Qty | Calories | Protein (g) | Fat (g) | Carbs (g) | Confidence | FatSecret Link | |---|-----------|-----|----------|-------------|---------|-----------|------------|----------------|

Confidence column:

• Copy EXACTLY from TABLE_ROW in server response
• Format: 88/100 ✅ or 65/100 ⚠️ or 35/100 ❌
• HIGH ✅ = exact or near-exact match
• MEDIUM ⚠️ = possible match — user confirmed
• LOW ❌ = weak match — user manually picked

FatSecret Link column:

• Copy EXACTLY from TABLE_ROW in server response
• Renders as: [🔗 View](https://foods.fatsecret.com/...)
• This is the live FatSecret URL for the user to verify independently

Nutrition values:

• Copy EXACTLY from TABLE_ROW in server response
• These are FatSecret API values — never substitute with Claude's estimates

---

Step 4 — Total Row

Copy TOTALS_ROW from get_day_summary response EXACTLY:

| **TOTAL** | — | — | — | — | **ΣCal** | **ΣP g** | **ΣF g** | **ΣC g** |

---

Step 5 — Percentage Row

Copy PCT_ROW from get_day_summary response EXACTLY:

| **% of 1650** | — | — | — | — | **XX%** | **P:XX%** | **F:XX%** | **C:XX%** |

Then on a new line below the table: Remaining: XXX kcal ✅ FatSecret REST API — Global Database (Live)

---

Step 6 — Warning

If get_day_summary response contains WARNING: line → copy it below the table as-is.

---

Step 7 — Date & History

• Same table maintained all day per person — new foods appended to bottom
• New day → fresh table
• Historical add R to DD-MM food: → Call get_diary(date) first → loads existing TABLE_ROWs → append new food

---

Step 8 — View Commands

"day R" / "summary R" / "calorie R today":
→ Call get_day_summary(today, 1650)
→ Build table using ALL_ITEMS_TABLE_ROWS from response
→ Show with TOTALS_ROW and PCT_ROW

"calorie R DD-MM":
→ Call get_diary(date)
→ Build table using TABLE_ROW entries from response

---

Step 9 — Delete

Delete triggers — any of these phrases mean the user wants to remove a food:

• "delete R coffee", "remove R coffee", "delete J rice"
• "delete R [entry_id]"
• "remove coffee from my diary"
• "I don't want coffee logged"

MANDATORY delete flow — always follow these exact steps:

IF user gives a food name (not an entry_id):
  Step A: Call FatSecret:get_diary { date: today }
          → Find the entry where food_entry_name matches the food name
          → Get its food_entry_id

  Step B: Confirm with user:
          "I found [food name] with entry_id [id] — delete it? (y/n)"

  Step C: If yes → Call FatSecret:delete_food_entry { entry_id: "[id]" }

IF user gives an entry_id directly:
  Step A: Call FatSecret:delete_food_entry { entry_id: "[id]" }

ALWAYS after deletion:
  Call FatSecret:get_day_summary { date: today, calorie_target: 1650 }
  Show updated table with the deleted row removed
  Show updated TOTAL and % rows

NEVER handle deletion by manually redrawing the table without calling delete_food_entry. NEVER say "the skill doesn't have a delete function" — it does, via FatSecret:delete_food_entry.

---

Step 10 — Update

"update R [entry_id] [new_qty] [meal]":
→ Call get_diary to get serving_id for that entry
→ Call update_food_entry(entry_id, serving_id, new_qty, meal)
→ Call get_day_summary to refresh
→ Show updated table

---

Step 11 — Weight

"weight R 72.5":
→ Call log_weight(72.5, today)
→ Reply: "⚖️ 72.5 kg logged for Rajiv on DD-MM"

---

Response Format

Every successful add:

1. "Calorie Table updated for [Rajiv/Jasleen] DD-MM"
2. The 9-column table (all values from FatSecret MCP)
3. Remaining: XXX kcal
4. ✅ FatSecret REST API — Global Database (Live)
5. Warning if applicable
6. Nothing else

---

Example

User types: add R 100 gm rice lunch

Skill does:

1. Person=Rajiv, Meal=lunch, Food=rice 100g
2. Calls search_multi_food(foods=["rice"], qty_map={"rice":"100g"}, region="IN")
3. Server returns TABLE_ROW: | White Rice (Cooked) | 100g | 91/100 ✅ | [🔗 View](https://foods.fatsecret.com/calories-nutrition/generic/white-rice-cooked) | 130 | 2.7 | 0.3 | 28.0 |
4. STATUS=AUTO_LOG → copies row directly into table, calls log_food
5. Calls get_day_summary → gets TOTALS_ROW and PCT_ROW
6. Shows:

Rajiv 15-06 Calorie Counter

| # | Food Name | Qty | Calories | Protein (g) | Fat (g) | Carbs (g) | Confidence | FatSecret Link | |---|-----------|-----|----------|-------------|---------|-----------|------------|----------------| | 1 | White Rice (Cooked) | 100g | 91/100 ✅ | 🔗 View | 130 | 2.7 | 0.3 | 28.0 | | | TOTAL | — | — | — | 130 | 2.7 | 0.3 | 28.0 | | | % of 1650 | — | — | — | 7.9% | P:8% | F:2% | C:90% |

Remaining: 1,520 kcal ✅ FatSecret REST API — Global Database (Live)