io.github.cyanheads/usda-mcp-server

@cyanheads/usda-mcp-server

Search foods, compare nutrients, and look up the full USDA FoodData Central database via MCP. STDIO or Streamable HTTP.

5 Tools • 2 Resources

Public Hosted Server: https://usda.caseyjhand.com/mcp

Overview

Nutrition data for roughly 400,000 foods from USDA FoodData Central, the US government's authoritative food composition database. Search by keyword or UPC, pull full nutrient profiles scaled to any portion, and compare foods side by side.

Tools

Resources

All resource data is also reachable via tools. Use usda_search_foods to discover FDC IDs before reading food resources.

Capability reference

usda_search_foods tool

• Query by keyword, ingredient name, or UPC/GTIN code; optional dataType[] filter across SR Legacy, Foundation, Survey (FNDDS), and Branded, defaulting to ["SR Legacy"] (or ["Branded"] once brandOwner is set)
• brandOwner and foodCategory filters narrow results; pageSize (max 50, default 10) and pageNumber paginate
• Returns FDC IDs plus a preview of key nutrients (energy, protein, fat, carbs) — not a complete profile
• Typed errors: query_empty (blank query), query_syntax (FDC rejected the query syntax), no_results (nothing matched in the requested data sources)

usda_get_food tool

• Full nutrient profile for one fdcId; optional nutrients[] filter to specific IDs
• Optional unit (g / oz / lb / kg / serving) with an optional quantity scales every value from the FDC per-100g basis; quantity defaults to 100 for mass units and 1 for serving, and unit="serving" multiplies the food's first defined portion weight by quantity
• Returns allPortions[] alongside the scaled result
• Typed errors: not_found (bad FDC ID), quantity_without_unit, no_portion_data (serving requested but the food has no portion data)

usda_get_foods tool

• Batch fetch for 2–20 FDC IDs in one call; values are per-100g only (no portion scaling)
• nutrients[] filter strongly recommended — full profiles for 20 foods are large
• Per-ID partial failure — IDs that return no data land in failed[] instead of aborting the batch

usda_compare_foods tool

• Side-by-side comparison for 2–5 FDC IDs, pivoted one row per nutrient and one column per food
• Defaults to 12 commonly compared nutrients (energy, protein, fat, saturated fat, carbs, fiber, total sugars, sodium, potassium, calcium, iron, vitamin C) unless nutrients[] is given
• All values scaled to a common quantity + unit basis (default 100 g; quantity is an amount in unit, so unit: "oz" alone compares per 100 oz)
• Proceeds with the valid foods when some IDs return no data; only throws too_few_foods when fewer than 2 resolve
• Rows where every food is null for that nutrient are filtered out

usda_list_nutrients tool

• Static reference table of ~150 tracked FDC nutrients — no API call
• Optional category filter: macronutrients, vitamins, minerals, lipids, amino_acids, other
• Returns each nutrient's ID, name, SR reference number, unit, and category — resolve a name to the ID used by nutrients[] elsewhere

usda://food/{fdcId} resource

• Same data as usda_get_food without portion scaling — full nutrient profile per 100g as application/json
• fdcId must be a bare positive integer (no sign, leading zero, decimal, or exponent); anything else throws invalid_id
• not_found when the ID doesn't exist in FDC

usda://nutrients resource

• Complete FDC nutrient reference — same content as usda_list_nutrients with no category filter
• Served as application/json with a 1-hour public cache hint, since the table is bundled at build time and identical for every caller

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

FoodData Central-specific:

• Type-safe client for the USDA FoodData Central REST API (api.nal.usda.gov/fdc/v1)
• Normalization layer that handles inconsistent API response shapes across search, single-food, and batch endpoints
• Batch API endpoint (/foods) with per-food partial failure reporting
• HTML response detection (rate-limit proxy returns 200 HTML) with service-unavailable error surfacing
• Static nutrient reference dictionary — ~150 nutrients with FDC IDs, SR numbers, units, and categories; no API call required

Agent-friendly output:

• Provenance preserved — FDC data source (dataType) on every food result so callers know whether they're reading curated research data or label-derived branded values
• Partial failure reporting — batch tools (usda_get_foods, usda_compare_foods) return successes alongside structured failed[] / missingData[] entries rather than failing the whole request
• Cross-reference hints — FDC IDs and tool names in descriptions so agents know exactly which call to make next (search → get → compare workflow)

Getting started

Public Hosted Instance

A public instance is available at https://usda.caseyjhand.com/mcp — no installation required. It serves Streamable HTTP; each client takes its own config shape.

For Claude Code:

claude mcp add --transport http usda-mcp-server https://usda.caseyjhand.com/mcp

For Cline (cline_mcp_settings.json):

{
  "mcpServers": {
    "usda-mcp-server": {
      "type": "streamableHttp",
      "url": "https://usda.caseyjhand.com/mcp"
    }
  }
}

For Cursor (mcp.json):

{
  "mcpServers": {
    "usda-mcp-server": {
      "url": "https://usda.caseyjhand.com/mcp"
    }
  }
}

Claude Desktop adds remote servers under Settings → Connectors.

Self-Hosted / Local

Add the following to your MCP client configuration file. See data.gov API key signup to generate a free API key.

{
  "mcpServers": {
    "usda-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/usda-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info",
        "USDA_FDC_API_KEY": "your-api-key"
      }
    }
  }
}

Or with npx (no Bun required):

{
  "mcpServers": {
    "usda-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@cyanheads/usda-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info",
        "USDA_FDC_API_KEY": "your-api-key"
      }
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "usda-mcp-server": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "MCP_TRANSPORT_TYPE=stdio",
        "-e", "USDA_FDC_API_KEY=your-api-key",
        "ghcr.io/cyanheads/usda-mcp-server:latest"
      ]
    }
  }
}

For Streamable HTTP, set the transport and start the server:

MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 USDA_FDC_API_KEY=your-api-key bun run start:http
# Server listens at http://localhost:3010/mcp

Prerequisites

• Bun v1.4.0 or higher (or Node.js v24+).
• A free USDA FDC API key — register at api.data.gov/signup. Required for food search, lookup, and comparison calls; usda_list_nutrients and the static nutrient reference work without it.

Installation

1. Clone the repository:

git clone https://github.com/cyanheads/usda-mcp-server.git

1. Navigate into the directory:

cd usda-mcp-server

1. Install dependencies:

bun install

1. Configure environment:

cp .env.example .env
# edit .env and set USDA_FDC_API_KEY

Configuration

See .env.example for the full list of optional overrides.

Running the server

Local development

• Build and run:

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:stdio
  # or
  bun run start:http

• Run checks and tests:

  bun run devcheck   # Lint, format, typecheck, security
  bun run test       # Vitest test suite
  bun run lint:mcp   # Validate MCP definitions against spec

Docker

docker build -t usda-mcp-server .
docker run --rm -e USDA_FDC_API_KEY=your-key -p 3010:3010 usda-mcp-server

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/usda-mcp-server. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for request-scoped logging, ctx.state for tenant-scoped storage
• Register new tools and resources via the barrels in src/mcp-server/*/definitions/index.ts
• Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.