Visual testing for Storybook, on your own infrastructure.

Up and running in three steps

1. Scaffold a server

   npm

   npx storyshelf server init

   pnpm

   pnpx storyshelf server init

   yarn

   yarn dlx storyshelf server init

2. Upload a Storybook build from CI

   npm

   npx storyshelf upload

   pnpm

   pnpx storyshelf upload

   yarn

   yarn dlx storyshelf upload

3. Review the diffs in your browser and approve or reject each change, or gate merges on the result.

Full walkthrough: Getting started.

What you get

A real review workspace

Filmstrip of changed stories, side-by-side, swipe, onion-skin and flip comparison, pixel-level zoom, and keyboard-driven approve/reject that jumps to the next change.

Branch-aware baselines

Baselines per branch with fallback to the default branch. Approve once, auto-approve on main, and keep long-lived branches with persistent labels.

Server-side capture

Upload your built Storybook; the server renders every story with Playwright (Chromium, Firefox, WebKit) or lightweight Puppeteer. Affected-only capture keeps CI fast.

Published Storybook

Share the latest build at /storybook or a per-project subdomain, gated to public branches.

Merge gate and notifications

Commit statuses for GitHub and GitLab, threaded comments, and Slack, Teams or email alerts when a build needs review.

AI triage and health (optional)

Bring your own model and get a verdict and per-snapshot notes on every build, plus a 30-day project health digest. Advisory only, off by default, secrets redacted.

Agent-ready (MCP)

An MCP server lets Claude Code, Cursor and other agents read builds, diffs and AI triage and comment on reviews. Local over stdio or shared over HTTP, scoped by your existing tokens, and it never approves for you.

Self-host anywhere

Docker Compose locally, or Terraform stacks for AWS, Azure and GCP. Bring your own database and storage; the database is also the capture queue.

AI that helps you review

Optional and advisory. Connect any model through the Vercel AI SDK (OpenAI, Anthropic, Google, Bedrock, Vertex, Azure, a gateway, or a local model) and StoryShelf will tell you whether a build’s changes look intended, need a human look, or look like a regression, and summarize how healthy each project is.

• Triage on the review page: a verdict, a short summary and a note per snapshot, generated on demand and cached.
• Project health: a score, verdict and trends on the Library page and as a badge on the project card.
• Under your control: off until a site admin enables a profile per project, a daily token budget across all instances, and prompts and screenshots never leave your chosen model or your traces.

AI triage above the review workspace: a verdict, a summary and notes per snapshot.

Agents can read all of it too: the MCP server exposes builds, diffs and triage to Claude Code, Cursor and friends.

Read about AI insights · Set it up · Try it with free models

Keep exploring

Getting started Scaffold a server, create a project, upload your first build.

Core concepts Projects, baselines, labels, builds, capture and retention: the mental model.

AI insights Optional AI triage and project health: setup, privacy, budget and API.

See it in action Screenshots of the projects, build and library views, plus the live demo.

API reference Every /api/v1 endpoint, grouped and searchable.

Chromatic comparison at a glance

	Chromatic (SaaS)	StoryShelf (self-hosted)
Pricing	5k snapshots free, then $179/mo per 35k	MIT, unlimited — infra cost only
Browsers	Chrome, Firefox, Safari, Edge (cloud)	Chromium/Firefox/WebKit per project + lightweight Puppeteer
Hosting	Chromatic cloud only	Your infra — Docker, Fly, Vercel, Lambda, Azure, GCP
Data	US/EU regions	Your DB/storage — full residency
Lock-in	Proprietary	Open source, standard /api/v1

Full breakdown and migration guide: Comparison · Migration.