Low-friction nutritional tracking plugin for Claude Code
npx claudepluginhub langadventurellc/health-ai-pluginLow-friction nutritional tracking. Describe what you ate and get calculated nutrition data with confidence scores.
A low-friction nutritional tracking system for Claude Code. Describe what you ate via text or photos and get back calculated nutritional information with confidence scores.
Two components work together:
The LLM reasons about what you ate and how much. The MCP server does the calculations - no nutritional math relies on LLM probability.
mise install # Install Node 24.11.0
npm install # Root: install git hooks (husky + lint-staged)
cd server
cp .env.example .env # Add your USDA API key (free from https://fdc.nal.usda.gov/api-key-signup)
npm install # Server: install project dependencies
Both npm install steps are required -- root installs repo-wide tooling (git hooks), server/ installs project dependencies.
mise run dev # Dev server with hot reload (from repo root)
# or
cd server && npm run dev # Equivalent
The server starts on http://localhost:3000 with a health check at /health and MCP endpoint at /mcp.
plugin/.mcp.json with your deployed server URL (replace PLACEHOLDER_URL)claude plugin add ./plugin
If the server has OAuth enabled (AUTH_ENABLED=true), Claude Code handles the OAuth 2.1 flow automatically on first connection. If auth is disabled, no authentication is needed.
mise run test # From repo root
# or
cd server && npm test # Equivalent
Pre-commit hooks automatically run ESLint, Prettier, type checking, and tests on staged files. To run manually:
mise run quality # Lint + format + type-check
mise run lint # ESLint --fix
mise run format # Prettier --write
mise run type-check # tsc --noEmit
The server deploys to AWS ECS Fargate via a GitHub Actions workflow (manual trigger). See infra/README.md for full setup instructions including Terraform provisioning, GitHub Actions variable configuration, and operational commands.
mise run deploy # Opens the deploy workflow in GitHub
In active development. The MCP server core is implemented with search_food, get_nutrition, calculate_meal, and save_food tools, SQLite caching, and USDA/Open Food Facts integration. See REQUIREMENTS.md for the full specification.
Implemented:
calculate_meal) with deterministic nutrient summing and coverage reportingsave_food) for restaurant items and nutrition labels, with 90-day TTL and upsert semanticsAUTH_ENABLED env var)nutrition-tracker skill (see plugin/README.md for setup)infra/README.md)This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.