crop-prices

A Claude Code plugin for US commodity crop prices. It bundles an MCP server that fetches prices for 37 crops from three free sources, and a skill that turns them into a one-screen market brief.

What you get

MCP tools

• list_commodities: crops and groups the server knows about.

• get_current_price: latest price for one crop (futures first, then farm, then global).

• get_price_history: a price series with start, end, low, high, and % change. Supports state-level farm prices (e.g. Iowa soybeans).

• compare_commodities: rank up to 12 crops by % change over the same period.

• get_group_snapshot: current prices for a whole group (grains, oilseeds, fiber, softs, specialty, forage, other).

Skill: crop-market-brief

Ask "give me a crop market brief" and Claude writes a short report: a headline, futures price tables by group, a 12-month trend table, and notes on units and how fresh each price is.

Related MCP server: USDA MCP Server

Install

Requirements: uv (the server runs with uv run).

/plugin marketplace add RootedInData/crop-prices
/plugin install crop-prices@crop-prices

USDA NASS API key (optional)

Futures and global prices work without a key. For USDA farm prices, get a free key at https://quickstats.nass.usda.gov/api and set it:

claude plugin configure crop-prices

The key is stored as a sensitive plugin option. For local development you can put it in .env instead (see .env.example).

Notes on the data

• Every price comes with its unit and source. Units differ across sources, even for the same crop (soybeans: cents/bushel futures, $/bushel farm price, $/metric ton global).

• Futures and farm prices measure different things; the skill never computes a % change between them.

• Prices are nominal (not adjusted for inflation).

• Not investment advice.

Development

uv sync
uv run pytest
uv run python -m server.crop_prices_server     # run the server over stdio
claude --plugin-dir .                          # load the plugin for one session

Self-hosting over HTTP (optional)

The plugin runs the server on your machine over stdio. server/http_app.py serves the same tools over Streamable HTTP, protected by a bearer token, for hosting.

Run locally

export MCP_AUTH_TOKEN=$(python3 -c "import secrets; print(secrets.token_urlsafe(32))")
uv run uvicorn server.http_app:app --port 8080
claude mcp add --transport http crop-prices-http http://localhost:8080/mcp \
  --header "Authorization: Bearer $MCP_AUTH_TOKEN"

Requests without the token get 401. If MCP_AUTH_TOKEN is unset, every request is refused.

Deploy to Google Cloud Run (needs a project with billing enabled; small personal use normally stays in the free tier, but Google Cloud has no hard spending cap)

PROJECT=your-project-id REGION=us-central1
gcloud services enable run.googleapis.com cloudbuild.googleapis.com \
  artifactregistry.googleapis.com secretmanager.googleapis.com --project $PROJECT
printf %s "$MCP_AUTH_TOKEN" | gcloud secrets create mcp-auth-token --data-file=- --project $PROJECT
printf %s "$NASS_API_KEY"   | gcloud secrets create nass-api-key   --data-file=- --project $PROJECT
# Let Cloud Run's default service account read the secrets
SA=$(gcloud projects describe $PROJECT --format="value(projectNumber)")-compute@developer.gserviceaccount.com
for s in mcp-auth-token nass-api-key; do
  gcloud secrets add-iam-policy-binding $s --member=serviceAccount:$SA \
    --role=roles/secretmanager.secretAccessor --project $PROJECT
done
gcloud run deploy crop-prices --source . --region $REGION --project $PROJECT \
  --max-instances 1 --memory 512Mi --allow-unauthenticated \
  --set-secrets MCP_AUTH_TOKEN=mcp-auth-token:latest,NASS_API_KEY=nass-api-key:latest

--allow-unauthenticated lets requests reach the app; the app itself checks the token. Then connect with claude mcp add --transport http crop-prices-cloud <service-url>/mcp --header "Authorization: Bearer $MCP_AUTH_TOKEN". Set a budget alert in the Billing console.

License

MIT