# modelfunction-cli

> Function-calling quality gate for agent tools. `mf` scans MCP tools and skills, lints description boundaries, and scores real models on a reviewed tool-selection suite. Local-first: no upload, no telemetry; the only outbound request is the model endpoint you configure.

## Use it when

- A model picks the wrong tool after another tool was added.
- A model refuses to call a tool because a required parameter looks unsourced.
- Two tools have overlapping descriptions and no mutual boundary.
- A function-calling score needs to become a CI regression gate.

## Quick start

```bash
npx -y modelfunction-cli@latest --version
npx -y modelfunction-cli@latest scan --root ./agent --out mf.suite.json
npx -y modelfunction-cli@latest lint --suite mf.suite.json
npx -y modelfunction-cli@latest audit --suite mf.suite.json --out mf-audit.md
```

## Agent entry points

- MCP server (stdio): `npx -y modelfunction-cli@latest mcp`
- Install gate: `import { check } from "modelfunction-cli"`
- Case authoring skill: `skills/mf-cases/SKILL.md` (shipped in the package)

## Docs

- [README](https://github.com/modelfunction/cli/blob/main/README.md): install, quickstart, data boundary.
- [Schema](https://github.com/modelfunction/cli/blob/main/cli/SCHEMA.md): suite, scorecard, lint, check() and MCP contracts.
- [Getting started](https://github.com/modelfunction/cli/blob/main/docs/getting-started.md): five steps to a first scorecard.
- [Local gate](https://github.com/modelfunction/cli/blob/main/docs/local-gate.md): pre-push, change gate and nightly regression.
- [Landing page](https://modelfunction.pages.dev/): product overview and example report.