gut's judgment calls as MCP tools, for Claude Code, Claude Desktop, Cursor and any other MCP client: likely, classify, rate and each, answered by a small, fast, cheap model, with YES, NO or UNSURE.

claude mcp add gut --env TYPESAFE_API_KEY=your-key -- uvx gutfeel-mcp

Or, in any client's JSON configuration:

{
  "mcpServers": {
    "gut": {
      "command": "uvx",
      "args": ["gutfeel-mcp"],
      "env": { "TYPESAFE_API_KEY": "your-key" }
    }
  }
}

The model comes from the environment: TypeSafe's Jev whenever TYPESAFE_API_KEY is set, or GUT_BACKEND for another -- openrouter, ollaya, ollama, openai, or zeroshot for a local model (uvx --with "gutfeel[local]" gutfeel-mcp).

Documentation: https://gutpy.dev/docs/mcp.html · The library: https://pypi.org/project/gutfeel/