CodeLoop Documentation
CodeLoop is simple.Three commands and your AI coding agent runs the build, the tests, the screenshots, the visual diffs, and the gate-check on every change — on its own — until the work is genuinely ready for review. The pages below are deep because the product has matured, but you don't need to read most of them to get value. Most users only ever touch this page and Quick Start.
- 1Sign in (one-time, opens browser)
npx codeloop auth - 2Move into your project folder
cd your-project # quote the name if it contains spaces, e.g. cd "Photometry DB" - 3Wire Cursor / Claude Code in this repo (one-time per repo)
npx codeloop init
Cursor users: press Cmd / Ctrl + Shift + J → Tools & MCP → toggle codeloop ON(one-time security consent — no deeplink works). Claude Code users: nothing to do (auto-authorized).
That's it. After init, your agent automatically runs verify → diagnose → fix → gate-check on every change. You write code, it ships green. Stuck? Run npx codeloop doctor. Full Quick Start →
The pages below are organised by intent. Pick the section that matches what you're trying to do; ignore the rest.
Start here
Read these and you can use CodeLoop. Most users never need anything else.
Quick Start
Get CodeLoop running in under 2 minutes. Three setup paths for every workflow.
Core concepts
The mental model behind CodeLoop — verify, diagnose, gate-check, runs, artifacts, gates, and the confidence score.
Architecture
How the pieces fit together end-to-end: MCP server, CLI, dashboard, GitHub Action, plugins, backend API.
API Keys
Generate via the browser auth flow, the CLI, or the dashboard. Rotate, revoke, and secure.
Agent setup
Pick the page that matches your editor. Each one walks through the simplest possible setup first.
Reference (look up when you need it)
You don’t need to read these to use CodeLoop. They’re here when you want to know exactly what a flag does.
Tool Reference
Complete reference for all 29 MCP tools — parameters, outputs, and examples.
CLI reference
Every command in the codeloop CLI — init, auth, verify, doctor, dashboard, design, baseline, configure.
Configuration
Full reference for .codeloop/config.json, plugins.json, figma.json, and every environment variable.
Glossary
Quick definitions for every term — verify, diagnose, gate, run, artifact, baseline, master spec, and more.
Capabilities (these run automatically — here’s how)
All of these run on their own once CodeLoop is initialised. Read these only when you want to tune behaviour or understand what the agent just did.
Visual review
Catch visual regressions automatically — capture, baseline, diff, and gate on a pixel-difference threshold.
Design compare (Figma)
Compare coded UI against Figma frames or local PNGs across viewports with a deterministic match score.
Recording & replay
Record video of real interactions, extract motion-validated key frames, and replay with correlated logs.
Multi-section orchestration
Section state machine, integration checks, replanning, and the 5-layer prompt manager.
Surfaces (optional places to view results)
Verification works without any of these. Open them when you want a richer view, a CI gate, or to extend CodeLoop.
Local Dashboard
Browse runs, screenshots, videos, and gate scores at http://localhost:3737.
GitHub Action
codeloop-verify@v1 — gate every PR with a confidence score and Verified by CodeLoop badge.
Plugin SDK
Wire any CLI test runner into the verify loop — Python/Django, Ruby/Rails, Go, custom scripts.
Self-host
Run the entire CodeLoop stack on your own infrastructure with one Docker Compose file.
Operations & policy
For when your security team, finance team, or platform team has questions.
Security & data handling
What leaves your machine, what stays local, key handling, telemetry, and self-host options.
Billing & usage
Pricing tiers, what counts as usage, the 14-day trial, the OSS plan, monthly vs annual.
Migration guide
Move from Cursor Bugbot, Devin, Chromatic, Percy, manual QA, or hand-rolled screenshot scripts.
Cross-OS Runbook
Install, configure, verify, and troubleshoot CodeLoop on macOS, Windows, and Linux.
Launch Operations
Pre-flight runbook for the public launch — Cursor Marketplace publish, Stripe LIVE flip, search-console verification, and launch-week post drafts.
Benchmark Harness
How CodeLoop benchmarks itself against Bugbot and vanilla Cursor — buggy-commits-50, the BENCHMARK_ENABLED CI gate, and the network-purity guarantee.
Tier-B Runbook (Windows)
Per-row checklist for the Test 24 sweeps that need a real Windows host — Flutter desktop, .NET / WPF / WinUI, Android, web, and multi-monitor.
Claude Code App Pack
5-app matrix you run inside Claude Code to validate CodeLoop's loop across Next.js, Flutter, native macOS, Express API, and Django + React stacks.
Troubleshooting
Symptom-first index — auth, init, verify, screenshot, video, gate, dashboard, and CI fixes.
FAQ
Frequently asked questions about setup, usage, billing, and troubleshooting.
Community
Help shape CodeLoop and earn free days while you do it.