# Loopcaptain

> An open protocol for agent loops and a local dashboard to monitor and steer them. A loop
> (a script, a Ralph loop, a long-running agent session) writes a few small files into
> .loops/<loop-id>/ in its own repository: loop.json, events.jsonl, inbox.jsonl. The board
> shows where each loop stands, what it cost and what waits for a person, and writes the
> person's answer, note or stop back. No dependency, no cloud, nothing leaves the machine.

## Start

- [Build a loop](standard/prompts/build-a-loop.md): Help me build an agent loop that reports to Loopcaptain.
- [Watch the loops on this machine](standard/prompts/watch-my-loops.md): Help me see and steer the agent loops on this machine with Loopcaptain.
- [Installation](docs/installation.md): one command, from the repository; Python 3.12 or later
- [Get started](docs/get-started.md): the example loop, free, in five minutes
- [Setup](docs/setup.md): what the board watches, checking a loop, when a loop does not show

## The standard

- [LOOPS.md](standard/LOOPS.md): the protocol, the helper to paste, conformance
- [AGENTS.md for a loop](standard/AGENTS.md): connect a loop, or build one from nothing
- [Works with every agent](docs/agents.md): how each coding agent adds the MCP server

## Optional

- [The command line](docs/reference/cli.md)
- [The MCP tools](docs/reference/mcp.md)
- [Security](SECURITY.md)
