Introduction

Bellows runs a coding agent against your repository on a cluster. Every run happens inside a container you define. It must pass the checks you define. Its result is a pull request you review.

Ways to use Bellows

Web app

Write tasks, answer the agent’s questions and review results in the browser.

CLI

bellows job create: script tasks from your terminal or CI.

MCP server

Drive Bellows from Claude Code, Claude.ai or any MCP client.

Cloud or self-hosted

	Bellows Cloud	Self-hosted
Setup	Sign in with GitHub, pick repos	Postgres, a GitHub App, Docker or Kubernetes
Runners	Managed	Your cluster (Helm chart)
Best for	Trying Bellows, most teams	Strict data residency, custom infra

These docs follow the Cloud path. To run Bellows yourself, see Self-hosting.

How it fits together

Part	What it does
Server	Stores tasks and serves the web app, the API and the MCP endpoint
Driver	Picks up queued jobs and starts a runner for each one
Runner	The container your repo’s `.bellows.yaml` describes. The agent works and the gates run here
Executor	The coding agent: Claude Code, OpenCode or a custom profile built on them

Overview

Use Bellows

Use cases

Configure your repo

Agents and integrations

CLI

MCP

Ways to use Bellows

Web app

CLI

MCP server

Cloud or self-hosted

How it fits together

Next steps

Quickstart

Core concepts

​Ways to use Bellows

Web app

CLI

MCP server

​Cloud or self-hosted

​How it fits together

​Next steps

Quickstart

Core concepts

Ways to use Bellows

Cloud or self-hosted

How it fits together

Next steps