MCP server de NutriRef

NutriRef

Pay-per-call USDA nutrition data for AI agents. Structured FoodData Central via the x402 micropayment protocol — agents pay $0.001–$0.005 in USDC per request, no signup, no API keys, no human auth flows.

Live at https://nutriref.xyz. Spec at /openapi.json · Swagger at /docs · Bazaar discovery at /.well-known/x402.

Endpoints

Method	Path	Price	Cache
GET	/v1/nutrition/search?q=&limit=	$0.001	24h
GET	/v1/nutrition/detail/{fdc_id}	$0.002	7d
POST	/v1/nutrition/compare	$0.003	derived
POST	/v1/nutrition/recipe	$0.005	derived

All values per 100g. Missing nutrients are null, not 0. compare returns per-nutrient winners (highest protein, lowest sodium, etc.). recipe scales by grams and sums.

Use it from Claude (or any MCP agent)

NutriRef ships an MCP server that exposes the four endpoints as native tools. Install it from PyPI:

pip install nutriref-mcp

Then add this to your MCP client config (Claude Desktop's claude_desktop_config.json, Claude Code's MCP settings, etc.):

{
  "mcpServers": {
    "nutriref": {
      "command": "nutriref-mcp",
      "env": {
        "PAYER_PRIVATE_KEY": "0x...your-funded-wallet-key...",
        "NUTRIREF_BASE_URL": "https://nutriref.xyz"
      }
    }
  }
}

Prefer not to install? Use uvx nutriref-mcp as the command to run it on demand. To work from a clone instead, pip install -e ".[mcp]" and set command to python with args: ["-m", "mcp_server"].

The wallet needs USDC on Base mainnet — gas is sponsored by the facilitator, so you only need stablecoin balance. The agent now has nutrition_search, nutrition_detail, nutrition_compare, nutrition_recipe and auto-pays per call.

Use it from any HTTP client

Unpaid requests get 402 Payment Required with x402 payment instructions. Any x402-aware client signs a gasless USDC authorization (EIP-3009) and retries automatically:

import asyncio
from eth_account import Account
from x402.client import x402Client
from x402.http.clients.httpx import wrapHttpxWithPayment
from x402.mechanisms.evm.exact import register_exact_evm_client

account = Account.from_key("0x...funded-wallet-key...")
client = x402Client(); register_exact_evm_client(client, account)

async def main():
    async with wrapHttpxWithPayment(client, base_url="https://nutriref.xyz") as http:
        r = await http.get("/v1/nutrition/detail/2012128")
        print(r.json())

asyncio.run(main())

Response example

GET /v1/nutrition/detail/173944:

{
  "fdc_id": 173944,
  "description": "Banana, raw",
  "data_type": "Foundation",
  "serving_size": 100, "serving_size_unit": "g",
  "calories": 89.0,  "protein": 1.1,    "fat": 0.3,
  "carbs": 22.8,     "fiber": 2.6,      "sugar": 12.2,
  "sodium": 1.0,     "cholesterol": null, "saturated_fat": 0.1,
  "vitamin_c": 8.7,  "calcium": 5.0,    "iron": 0.3,  "potassium": 358.0
}

---

Self-hosting

NutriRef is open source; the live instance at nutriref.xyz is one deployment among many possible. To run your own:

cp .env.example .env
# fill in USDA_API_KEY (free at https://fdc.nal.usda.gov/api-key-signup.html)
# and X402_RECEIVER_ADDRESS (an EVM address that should receive payments)
docker compose up --build
curl http://localhost:8000/health

Configuration

Var	Required	Default	Purpose
USDA_API_KEY	yes	—	Free key from fdc.nal.usda.gov
USDA_BASE_URL	no	https://api.nal.usda.gov/fdc/v1
REDIS_URL	no	redis://redis:6379/0	Response cache
X402_NETWORK	no	base-sepolia	base for mainnet
X402_RECEIVER_ADDRESS	yes	—	EVM address that receives USDC
X402_FACILITATOR_URL	no	https://x402.org/facilitator	https://api.cdp.coinbase.com for mainnet
CDP_API_KEY_ID	mainnet only	—	Coinbase Developer Platform key ID
CDP_API_KEY_SECRET	mainnet only	—	Coinbase Developer Platform key secret
LOG_LEVEL	no	INFO

For mainnet you need a Coinbase CDP account and the public x402 facilitator at https://api.cdp.coinbase.com. Testnet works for free with the community facilitator at https://x402.org/facilitator.

Architecture

agent → x402 middleware → route handler → cache (Redis) → USDA FDC API

search and detail cache USDA responses directly. compare and recipe compose from the cached detail data — no extra USDA calls when warm. The cache is a meaningful cost lever: warm requests return in <50ms and never hit USDA.

Tests

pip install -e ".[dev]"
pytest

Example: Claude agent that uses NutriRef

examples/meal-planner/ is a complete, ~150-line agent that gives Claude the four NutriRef endpoints as tools and asks it to plan a day of meals hitting a calorie/protein goal. Worth reading if you're wiring NutriRef into your own agent — the tool schemas and the payment loop are all there. See examples/meal-planner/README.md.

Repo layout

app/                # FastAPI service
  main.py             # app factory + x402 init
  routes/             # search, detail, compare, recipe
  landing.py          # / (public landing page)
  discovery.py        # /.well-known/x402, /llms.txt, /.well-known/ai-plugin.json, /logo.svg
  usda.py             # async USDA client
  cache.py            # Redis wrapper
  normalize.py        # USDA → flat 13-nutrient schema
mcp_server/         # MCP server wrapper for agent use
examples/           # worked agent examples (meal planner)
scripts/            # CDP wallet bootstrap + payer-side test
tests/              # pytest + respx + fakeredis

Acknowledgments

• USDA FoodData Central for the data.
• x402 for the payment protocol.
• fastapi-x402 for the server middleware (with a small EIP-712 patch we apply at startup for Base mainnet USDC).