ag-bids-mcp

Model Context Protocol server exposing live + historical commodity price data collected by ag-monitor. Sits behind the user's MetaMCP gateway at https://mcp.jpaul.io/metamcp/ag-bids/mcp.

What it does

Lets an LLM client (Claude Desktop, OpenWebUI, anything that speaks MCP) ask plain-English questions like:

• "What's the best place to sell corn today?"
• "What's the current price of lime?"
• "Show me the corn basis trend at Andersons Greenville over the last 30 days."
• "Which sources are stale or failing?"

How it talks to data

All data is read from ag-monitor over HTTPS:

ag-bids-mcp ── X-API-Key ─► https://agbids.paul.farm/api/data/*

Endpoints used: /api/data/latest, /history, /best, /inputs, /sources, /deliveries. See the ag-monitor source for the contract.

Authentication

No in-container auth. The MCP's port 8000 is never exposed outside the private mcp-servers_mcp Docker network on .0.2. The only client that can reach it is MetaMCP, and MetaMCP enforces auth at the gateway → client edge (bearer token / OAuth in its UI).

This matches the zerto-docs-mcp pattern.

Local dev (stdio)

python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # fill in AG_BIDS_API_KEY + AG_BIDS_MCP_USER/PASS
MCP_TRANSPORT=stdio python -m ag_bids_mcp.server

Wire into Claude Desktop's claude_desktop_config.json:

{
  "mcpServers": {
    "ag-bids": {
      "command": "/path/to/venv/bin/python",
      "args": ["-m", "ag_bids_mcp.server"],
      "env": {
        "MCP_TRANSPORT": "stdio",
        "AG_BIDS_API_URL": "https://agbids.paul.farm",
        "AG_BIDS_API_KEY": "..."
      }
    }
  }
}

Deploy (MetaMCP host)

See deploy/README.md. Container image is pulled from git.jpaul.io/justin/ag-bids-mcp:latest; Watchtower on the MetaMCP host auto-updates it every 5 minutes.

Tools exposed

Tool	Returns
best_local_bid(commodity)	Where to sell corn, soy, or wheat for this month's delivery — markdown one-liner + table
current_lime_price()	Latest lime quotes across all manual-entry sources
current_input_price(product?)	MAP / Potash / Lime — all three or one
latest_prices(commodity?, source?, delivery?)	Snapshot table, optionally filtered
price_history(commodity, source?, delivery?, days?)	Compact time series
list_sources()	Active scrapers + last-success timestamps
list_commodities()	corn, soy, wheat, map, potash, lime
list_deliveries(commodity)	Posted delivery labels, chronological
source_health()	Stale / failing / healthy summary
todays_summary()	Same shape as the morning brief snapshot