bestpred-mcp

USDA's best prediction of dairy lactation yields (bestpred, the Python port of BESTPRED 2.0rc7) as tools any AI agent can call. An agent can't work out a lactation record in its head; with this it runs the method USDA uses for official records, and reports the numbers it gets back.

It is an MCP server. MCP is the open standard agents use for tools, so the same server works in Claude, ChatGPT and Codex, Cursor, VS Code and Gemini CLI.

Install

You need uv (it runs the server; nothing else to install).

uvx bestpred-mcp

• Claude Code: claude mcp add bestpred -- uvx bestpred-mcp

• Claude Desktop, Cursor, Gemini CLI: add to the app's MCP settings file:

  { "mcpServers": { "bestpred": { "command": "uvx", "args": ["bestpred-mcp"] } } }
  

• VS Code (.vscode/mcp.json): { "servers": { "bestpred": { "command": "uvx", "args": ["bestpred-mcp"] } } }

• Codex plugin (with the skill): codex plugin marketplace add padge01/bestpred-python, then install BESTPRED from the Plugins tab.

• Codex, the server alone (~/.codex/config.toml):

  [mcp_servers.bestpred]
  command = "uvx"
  args = ["bestpred-mcp"]
  

Then ask your agent, for example: "Work out 305-day milk for every cow in tests.csv".

The tools

Pounds by default; pass units: "kg" for kilograms. With no herd average given, the breed average is used.

What stays on your computer

Run this way, the server runs on your computer: your files are read there and the table is written there. What your agent shows the AI model (the summary it reads back) goes to that model's company, as everything in a chat does.

bestpred-mcp --http serves the same tools over HTTP for hosting. Hosted, it reads no files, keeps nothing, and takes at most 5,000 lactations a call.

Licence

Apache-2.0. BESTPRED is USDA-ARS public-domain work (VanRaden; Cole and VanRaden); this is not affiliated with or endorsed by USDA. See the main repository's NOTICE.