# fagan

An autonomous pipeline where a frontier model plans and reviews while a local model writes the code. Runs on macOS; on Linux it goes through Ollama or LM Studio.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed · ★ 5
- Author: motock
- License: Apache-2.0 · policy=link-only · description only; the file stays with the author
- Source: https://github.com/motock/fagan
- Updated: 2026-10-24
- HTML: https://affpapa.org/en/nevibecode/mcp/fagan

## What it does

fagan splits engineering work by cost. A frontier model breaks the task down, writes the plan, reads the diff and settles risky calls. A local model produces the implementation without burning tokens. The trust comes from gates rather than from the author: tests are written before the code, review runs as its own pass, and anything irreversible pauses for a human.

What you install is a Python MCP server, a dashboard, and a set of persona subagents plus a decision policy file. Setup builds a `.venv`, pulls the requirements, and you register the server in Claude with `claude mcp add`. The pipeline shells out to `git`, `gh` and the `claude` CLI; `ollama` and `docker` matter only on the local route. The scheduler and MLX supervisor ship as `launchd` jobs, so on Linux you render systemd units with `scripts/generate_systemd_units.sh` or run the entry points in a terminal session.

A first run needs no local model. Until you configure one, implementation and review go to the `claude` backend, which calls the Claude CLI. That is where you begin, not where you want to end up — the cost split only kicks in once the implementation role is pointed at a local model on purpose. Windows is untested.

## Requirements
- Needs: Python and uv (commands run via uvx)

## Install

### Claude Code (native platform)

**Install**
- The MCP registry has no launch command for it: see the repository README

## When to use it
- break a large task into stages and order the dependencies
- write tests before the implementation and keep the red phase honest
- run an independent review pass over the diff before merging
- route routine implementation to a local model instead of paying per token
- hold irreversible actions until a human confirms

## Example prompts
- Break this task into stages and draft a plan in fagan
- Send the implementation to the local model and keep review on the frontier model
- Show me the story status on the fagan board

## Similar
- [rpi](https://affpapa.org/en/nevibecode/skills/rpi.md) — skill · 1500 installs
- [Pair Programming](https://affpapa.org/en/nevibecode/skills/pair-programming.md) — skill · 1200 installs
- [clean-code](https://affpapa.org/en/nevibecode/skills/sickn33-clean-code.md) — skill · 12400 installs
- [tdd-refactor](https://affpapa.org/en/nevibecode/commands/tdd-refactor.md) — command · ★ 41823
- [evanflow](https://affpapa.org/en/nevibecode/plugins/evanflow.md) — plugin · ★ 419
- [draft](https://affpapa.org/en/nevibecode/plugins/draft.md) — plugin · ★ 40

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md