Artifact Server: A Self-Hosted Home for the Files Your AI Agents Keep Scattering

Artifact Server is an open source, self-hostable hub where AI agents publish their generated designs and code, and where teams review, comment on, and version that work on their own infrastructure. The video is a 33-second screen recording that scrolls through the GitHub README for plannotator/artifact-server, a project the account spotted on Hacker News. The pitch lands because the problem is familiar to anyone running coding agents daily. Agents produce HTML mockups, diagrams, prototypes, and review docs faster than any team can organize them, and those files end up strewn across chat threads, temp folders, and third-party artifact viewers.

The Problem It Targets

The README section shown in the frames, titled "Why Artifact Server," states it plainly. Teams now use HTML, images, and video for plans, mockups, prototypes, and code reviews, and that work quickly becomes disorganized while providers try to lock it inside their platforms. The narration frames the fix the same way. Give developers and AI agents one secure place to publish, review, and version their work, stored by project with comments and sharing built in.

Positioned Against Claude Code Artifacts

The repo's own tagline, visible in the first frames, calls it "the open-source, self-hostable alternative to Claude Code artifacts." That positioning matters. Anthropic's artifact viewing lives inside Anthropic's platform. Artifact Server lets you keep the same publish-and-review workflow on hardware you control, under an AGPL-3.0 license shown in the repo's tab bar.

MCP Is the Connection Layer

The feature the narrator calls the standout is the built-in MCP server. A diagram in the README shows agents like Claude, Codex, Cursor, Copilot, Pi, and OpenCode connecting through MCP to publish and update versioned artifacts. Setup as shown on screen takes two commands, artifactserver connect and npx skills add plannotator/artifact-server, after which you can ask your agent to upload a file and it returns a full-screen review link. The README also shows live session delivery for Pi and Claude Code Channels, with the OpenCode adapter marked as not yet release-qualified.

Deployment Options Cover Solo to Team

The README frames show a range of deployment targets. Run it on a laptop for one developer, or deploy it for a private team on Cloudflare, a single server, Kubernetes, AWS, or Google Cloud. Cloudflare is called out as the main hosted deployment target. The self-hosting angle is the core of the ownership claim in the narration, since your artifacts and review data stay on infrastructure you run.

Who Built It

The project comes from Plannotator, whose branding appears on the README banner alongside a promo for Plannotator Workspaces, described as "the planning layer for your software factory." Artifact Server appears to be the open source piece of that larger product line, which is worth knowing when evaluating how the free tool and the commercial offering relate.

Key Takeaways

• Artifact Server gives AI agents and teams one self-hosted place to publish, review, comment on, and version generated artifacts.
• It positions itself as the open source, self-hostable alternative to Claude Code artifacts, licensed AGPL-3.0.
• A built-in MCP server connects agents like Claude, Codex, Cursor, Copilot, Pi, and OpenCode, with a two-command setup.
• Deployment scales from a single laptop to Cloudflare, Kubernetes, AWS, or Google Cloud, with Cloudflare as the main hosted target.
• It is built by Plannotator and sits alongside their commercial Workspaces product.

Resources

• plannotator/artifact-server on GitHub