artifact-hub
Enables MCP clients to publish pages and design canvases to a self-hosted Artifact Hub, participate in pinned review comment threads, manage versions, share artifacts, and work with canvases and clickable prototypes.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Artifact Hub
A self-hosted home for what your AI tools build. Any MCP client, from Claude to an agent you wrote yourself, publishes pages and design canvases to the hub. You and your team open them in the browser and comment on the page. The AI reads each comment and replies. Everything is stored in one SQLite file on your own server.
Landing page · Reference · MIT license

What you get
A review loop the AI takes part in. Comments are pinned to an element on the page. The AI reads them, replies, fixes the page and resolves the thread.
Versions. Every change is saved as a new version. You can compare versions and go back.
A canvas for boards. Pages sit side by side on an infinite wall. Focus one, present them in order, or edit in place with layers and properties.
Click dummies. Wire boards into a clickable prototype with plain attributes or the editor's Prototype panel. Hotspots go to other boards with a transition and a back stack. Components have states for hover, press, focus and open, and selects, radios and checkboxes can drive them. Actions can depend on a field (
go:Cart.html if #email:valid & #terms:checked). Dialogs, drawers, sheets and popovers open over a board, and a flow view draws the links between boards. Try it in play or present mode.Sharing. View-only or comment links for guests, who need no account. Members have the role owner, editor or commenter.
Control over the AI. When you connect a tool, you choose what it may do: read, comment, publish and edit, or share. It never gets more rights than you have.
Extras. Interactive pages that store data, workspaces with files shared between boards, PNG and SVG export, and
.zipexport and import.
Related MCP server: Stela
Quick start
You need Docker, or Node 24 or newer.
docker run -d --name artifact-hub -p 3000:3000 -v artifact-hub:/data \
-e NUXT_BASE_URL=https://hub.example.com \
ghcr.io/czepter/artifact-hub:latestWith Docker Compose, use the docker-compose.yml in this repo. To run from source:
git clone https://github.com/czepter/artifact-hub.git
cd artifact-hub
npm install
npm run dev # http://localhost:3000Open the hub and create the owner account on the sign-in page. The first account is the owner. After that, people join by invitation: the owner creates an invite link under Members and sends it.
For real use, serve the hub over HTTPS and set NUXT_BASE_URL to its public address. Passkeys only work over HTTPS or on localhost.
Connect an AI tool
Add https://<your host>/mcp to your AI tool as an MCP server. In Claude this is a custom connector. The tool registers itself,
sends you to the hub to sign in, and shows a consent screen where you pick what it may do. The hub has a guide tool that tells the AI how to structure
artifacts, so you can start by asking it to publish something.
A grant belongs to the member who approved it. You can review and revoke grants under Account, and the owner sees all of them under Admin.
Configuration
All settings are environment variables. None are required to try it out.
Variable | What it does |
| The public address, for example |
| Signs sessions. If unset, the hub generates one and stores it next to the database in |
| Turn on "Sign in with GitHub". Both are needed. The callback URL is |
| Path of the SQLite database file. By default the hub creates it in a |
To back up the hub, copy the database file and auth-secret from the data directory (the /data volume in Docker).
To update, pull the new image and restart the container. The data stays in the volume.
Boards are drawn as text outlines unless the server has a browser to render them. For real pictures in PNG and SVG exports, build the image with
Chromium: docker build --build-arg WITH_CHROMIUM=1 . (details in the reference).
To deploy on Coolify, use docker-compose.coolify.yaml.
Development
npm install
npm run dev:login # dev server that signs every request in, no account needed
npm run test:unit # canvas, geometry, HTML-edit and rendering testsdev:login accepts the bearer token devtoken on the API and on /mcp, which is handy for scripts and tests. It only works in
nuxt dev. A production build ignores it and refuses to start if NUXT_DEV_LOGIN_EMAIL or NUXT_DEV_TOKEN is set, so never set them on a deployed instance.
The API tests in tests/*.mjs run against a dev-login server. For example, start PORT=3111 npm run dev:login, then run node tests/workspaces-test.mjs.
node scripts/seed-demo.mjs publishes a sample design canvas with a comment thread to a running hub.
The stack is Nuxt 4, shadcn-vue on reka-ui, Tailwind v4 and SQLite (Node's built-in node:sqlite). See CONTRIBUTING.md before opening a pull request.
Known gaps
There is no rate limiting on
/oauth/registerand/oauth/token. Put a limit in front of them in your reverse proxy.Invitations are links, not e-mails. There is no password reset by e-mail and no e-mail verification.
Guests who comment give an e-mail address that is not verified.
Templates for documents, slides and dashboards do not exist yet. The canvas is the only structured type.
Behind some reverse proxies the live updates need response buffering turned off for
/api/view/*/events. This has not been tested through a real proxy.
The full list is at the end of the reference.
Documentation
Reference: MCP tools, workspaces, shared files, rendering, admin, interactive pages, the viewer and editor, import.
PRODUCT.md and DESIGN.md: what the product is for and how the interface looks.
Canvas upgrade plan: design notes behind the canvas editor.
SECURITY.md: how to report a vulnerability.
License
MIT. shared/vendor/open-design/ is copied from nexu-io/open-design (Apache-2.0), see its NOTICE.md.
Artifact Hub is an independent project and is not affiliated with Anthropic.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
MarkupBase turns AI-generated Markdown and HTML into durable, versioned artifacts that people can review and discuss. Its MCP server lets agents publish new versions, preserve contextual comments, include hosted images, and respond to feedback through secure account-linked identities, creating a clear human review boundary without requiring real-time editing.
Self-hosted AI workspace: semantic RAG document search, long-term memory, and AI chat via MCP.
Publish websites with a small backend (KV, SQLite, files). MCP for AI agents.
AI-native git hosting — repos, PRs, issues, CI gates, and AI code review over MCP (60 tools).
Related MCP Servers
- FlicenseAqualityDmaintenanceSelf-hosted enterprise knowledge hub that compiles organizational docs into a structured wiki and serves it to AI clients via MCP with fine-grained access control.11,522-
- AlicenseNot gradedqualityDmaintenanceSelf-hosted MCP server for publishing, sharing, and reviewing AI agent artifacts with version history and pinned comments.MIT
- AlicenseNot gradedqualityBmaintenanceA local MCP server that lets AI agents publish markdown/HTML to a browser viewer with session management, revision diffs, and live updates via a single tool call.2MIT
- AlicenseNot gradedqualityCmaintenanceA machine-local, agent-writable knowledge base MCP server that lets coding agents autonomously record and retrieve hard-won operational facts, incidents, and corrections as markdown pages in a git repo, with provenance, talk pages, and search.5 npmMIT