Local-first nutrition tracking MCP server for Claude Desktop. Log meals by chatting with Claude, see macros in a visual dashboard. No cloud, no account, no subscription.
Pre-alpha. Expect rough edges and breaking changes. Feedback welcome.
- Log meals by chatting. "I had two eggs and a slice of toast" — Claude looks up the foods, picks portions, writes them to your local database.
- Visual dashboard. Calories, protein, carbs, fat, fiber against your goals. Renders as an in-chat artifact that updates in place — no window flipping.
- Barcode lookup. Read the barcode off a package, get the entry.
- Label photos. Snap a nutrition label, Claude reads it natively.
- Goals you set once. Calories and macro targets stored locally.
- Your data stays yours. Everything lives in
~/.digest/digest.dbon your machine. SQLite. Export to CSV anytime.
No meal plans. No grocery lists. No social feed. No streaks or gamification. No coaching nudges. No account. No cloud. No ads. No subscription.
It's a log and a dashboard. That's the whole product.
- USDA FoodData Central — 380K+ generic and branded foods, public domain
- Open Food Facts — 4M+ products, CC0, great for barcodes and international brands
Both are queried together and results deduplicated.
pip install digest-mcpAdd to your claude_desktop_config.json:
{
"mcpServers": {
"digest": {
"command": "digest-mcp"
}
}
}Restart Claude Desktop.
| Variable | Description |
|---|---|
USDA_API_KEY |
Free key from fdc.nal.usda.gov. Without it, falls back to a rate-limited demo key. |
DIGEST_DB_PATH |
Custom path for the SQLite database. Default: ~/.digest/digest.db |
Open a chat and talk normally:
"Log 3 scrambled eggs and 2 slices of whole wheat toast for breakfast"
"Set my goals to 2400 cal, 180g protein, 250g carbs, 80g fat"
"Show me today's dashboard"
"How much protein have I had today?"
"Look up barcode 049000042566"
"Delete the second entry from lunch"
search_food, lookup_barcode, log_meal, update_entry, delete_entry, get_food_log, export_log, set_goals, get_goals, daily_summary, generate_dashboard
~/.digest/digest.db — a single SQLite file. Back it up, copy it, inspect it, delete it. It's yours. The only network calls are food lookups to USDA and Open Food Facts.
MIT
Issues and PRs welcome. This is early software — the fastest way to help is to try it and report what breaks.