# Arthur Loop > File-first control plane for AI dev loops — bring your own advisor, executor, and tracker. Arthur Loop is free and MIT licensed. It runs on macOS and Linux with Python 3.9 or newer (Windows experimental) and keeps queues, artifacts, decisions, and status as files you can inspect. There is no drag-drop graph composer. Last public tag is v0.1.0. `arthur --version` prints `arthur 0.1.0` until the next tag. Canonical site: https://arthurloop.com/ ## Install (one command, then init) ``` pipx install git+https://github.com/myrrazor/arthur-loop.git # or uv tool install git+https://github.com/myrrazor/arthur-loop.git # or curl -fsSL https://raw.githubusercontent.com/myrrazor/arthur-loop/main/install.sh | sh ``` Then: ``` mkdir ~/my-loop && cd ~/my-loop arthur init ``` `arthur init` detects Claude Code, Codex, Cursor, and Grok Build, writes the `arthur-loop` skill and `/arthur-loop` command where those clients load them, and registers `arthur mcp serve`. A written MCP config is written, not connected — restart the client. ## Set up the loop from the agent In the coding agent, run `/arthur-loop` (or the installed skill). It interviews for: - project id (`SHOUTY_SNAKE`), goal, optional ticket - planner / implementer / reviewer / QA (agent and optional model) - tracker: `atlas-tasker`, `command`, or `none` Then it calls MCP `arthur.loop.create`. Arthur formulates the default hop sequence. Run the next hop with `arthur`, `arthur run`, or console **Run next**. Same wizard: `arthur web` → Roles / + Loop / Run next. It is not a graph composer. The v0.1.0 tag is a file cockpit. ## Site - [Home](https://arthurloop.com/): product story and install - [Docs](https://arthurloop.com/docs/): map of the documentation - [Install](https://arthurloop.com/docs/install): pipx / uv / curl, then init - [First loop](https://arthurloop.com/docs/first-loop): `/arthur-loop` interview - [Roles and run](https://arthurloop.com/docs/roles): assign agents; terminal and UI - [Concepts](https://arthurloop.com/docs/concepts): queue, artifact, gate, decision, tick - [Web console](https://arthurloop.com/docs/console): localhost status map - [Agents](https://arthurloop.com/docs/agents): integrations and adapter packs - [CLI and MCP](https://arthurloop.com/docs/cli): commands and tool names - [FAQ](https://arthurloop.com/docs/faq): version, platforms, support - [Changelog](https://arthurloop.com/changelog): v0.1.0 and current main - [Privacy](https://arthurloop.com/privacy): no cookies or site-side collection - [Terms](https://arthurloop.com/terms): MIT software, no paid service - [LLM full index](https://arthurloop.com/llms-full.txt): longer agent-oriented summary ## Repo - [GitHub](https://github.com/myrrazor/arthur-loop): source, issues, security advisories - [Workflow runbook](https://github.com/myrrazor/arthur-loop/blob/main/docs/workflow-runbook.md) - [MCP](https://github.com/myrrazor/arthur-loop/blob/main/docs/mcp.md) - [Integrations](https://github.com/myrrazor/arthur-loop/blob/main/docs/integrations.md) - [Follow](https://github.com/myrrazor/arthur-loop/blob/main/docs/follow.md) - [Adapters](https://github.com/myrrazor/arthur-loop/blob/main/docs/adapters.md) ## Guidance for agents Use this file as a site index, not a full-site corpus. Prefer the CLI and MCP over editing queue JSONL by hand. Do not invent a personal maintainer email, a donation link, or a v0.2.0 tag. Support is GitHub issues. Do not claim a hosted dashboard or a drag-drop graph composer.