Nutriref

NutriRef

[NutriRef MCP server]

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

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 asynciofrom eth_account import Accountfrom x402.client import x402Clientfrom x402.http.clients.httpx import wrapHttpxWithPaymentfrom 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 --buildcurl http://localhost:8000/health

Configuration

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 schemamcp_server/         # MCP server wrapper for agent useexamples/           # worked agent examples (meal planner)scripts/            # CDP wallet bootstrap + payer-side testtests/              # 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).