NutriClarity

CI PyPI Python License: MIT

An open-source Model Context Protocol server that gives any LLM access to food nutrition data — look up products by barcode, search by name, read Nutri-Score / NOVA / Eco-Score ratings, and compare products side by side.

Powered by Open Food Facts, a free, open, crowd-sourced food products database. No API key required.

Tools

Reading needs no API key or account — it's open data. Only the optional write tool requires an Open Food Facts login.

Quick start

The easiest way to run it is with uvx (no install, no clone):

uvx nutriclarity-mcp

Claude Desktop / Cursor / any MCP client

Add this to your MCP config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "nutriclarity": {
      "command": "uvx",
      "args": ["nutriclarity-mcp"]
    }
  }
}

Then ask things like:

• "What's the nutrition of barcode 3017620422003?"
• "Search for oat milk and show me the healthiest option."
• "Compare Coke and Pepsi nutrition."

Docker

docker run -i --rm ghcr.io/nitishkp001/nutriclarity-mcp

MCP config:

{
  "mcpServers": {
    "nutriclarity": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "ghcr.io/nitishkp001/nutriclarity-mcp"]
    }
  }
}

Optional: write support (adding/editing products)

By default the server is read-only — no account, no risk. If you want the LLM to be able to contribute or correct product data, enable the add_or_update_product tool by providing an Open Food Facts account:

Writes go to the sandbox by default, so you can test safely without touching the real database. Set OFF_ENVIRONMENT=production only when you deliberately want to edit the live, public database. Sandbox accounts are separate from production accounts — register on each site you target.

{
  "mcpServers": {
    "nutriclarity": {
      "command": "uvx",
      "args": ["nutriclarity-mcp"],
      "env": {
        "OFF_USERNAME": "your-username",
        "OFF_PASSWORD": "your-password",
        "OFF_ENVIRONMENT": "sandbox"
      }
    }
  }
}

⚠️ Open Food Facts is a shared, public database used by millions. If you enable production writes, make sure the data you submit is accurate and taken from the real product label. The tool changes only the fields you pass.

Local development

Requires uv.

git clone https://github.com/nitishkp001/nutriclarity-mcp
cd nutriclarity-mcp
uv sync --extra dev        # install deps
uv run nutriclarity-mcp    # run the server over stdio
uv run pytest              # run tests
uv run ruff check .        # lint

To inspect the tools interactively, use the MCP Inspector:

npx @modelcontextprotocol/inspector uv run nutriclarity-mcp

How it works

• Built with FastMCP.
• Calls the Open Food Facts REST API (/api/v2/product/{barcode} and /cgi/search.pl).
• Sends a descriptive User-Agent as Open Food Facts requests, and asks only for the fields it needs to keep responses small.

Data & attribution

Product data comes from Open Food Facts and is made available under the Open Database License (ODbL). Individual contents are under the Database Contents License. Nutrition data is crowd-sourced and may be incomplete or inaccurate — do not rely on it for medical decisions.

License

MIT © 2026 — see LICENSE. This project is not affiliated with or endorsed by Open Food Facts.