SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI productivityartifact-site

artifact-site

One place for every AI agent's pages and docs: publish HTML, build folders, ZIPs and PDFs as sandboxed, versioned share links, then search, read and update them. Hosted or self-hosted; remote MCP

Open website

About this tool

What is artifact-site artifact-site is a self-hosted workspace that provides individuals and teams with a central place to store pages and documents generated by AI agents, positioned similarly to a self-hosted version of Claude Artifacts or OpenAI Sites. There are currently many AI agents to choose from, but their outputs are often scattered across local computers, chat histories, and cloud platforms, making them difficult to organize, share, and review. artifact-site brings together the work of various agents, turning pages and documents into links that can be shared, commented on, access-controlled, and have version history.

Main capabilities

  • Centralized management: organize pages and documents with folders, and find them through full-text search (including Chinese). Team members and agents can only find content they have permission to access.
  • One-click publishing and sharing: upload HTML, a build folder, a ZIP, or a document to get a link. Large sites support chunked uploads. You can share with everyone, logged-in users, specified people, or people who have a password.
  • Continuous improvement: edit HTML text or source code in the browser, or let an agent update the entire site. Every change keeps a version, supporting rollback and saving a copy.
  • Agent collaboration: publish and update via CLI or MCP, search by content, and read extractable text. For example, find a previous report, update it at the same address, and make it available for the team to view.
  • Publishing types: HTML, build folders, ZIP, and PDF files can be published. Web projects must first be built into static files before uploading; the platform does not run application backends or build tasks. PDF and Office documents do not support visual HTML editing, and scanned images are not automatically OCR'd.

How to use Cloud version: hosted by Lexmount, no deployment required. Drag and drop files to publish and get a shareable link. You can upload without logging in, or log in with Google or GitHub to keep sites, organize them with folders, and control access permissions for each link. Anonymous uploads expire after a few days. Local deployment: requires Git, Make, Bash, Docker 24+, and Compose plugin 2.24+. On macOS use Docker Desktop; on Windows run in WSL2 and enable Docker Desktop's WSL integration. No need to install Node, a domain, or an identity provider. In a freshly cloned repository, run three commands (the first time it will downloaddependencies and build the image, then start the application and Postgres), open http://127.0.0.1:4300，拖入 HTML, a static site folder, or PDF to view it. New sites are private by default; change the access permission of the main link in Sharing before sharing. The service is accessible only locally by default; make down stops it and preserves data. Connect a coding agent: three entry points are available.

  • Skill: npx skills add lexmount/artifact-site, install the agent skill and give the server address to the agent.
  • CLI: npm install -g @artifact-site/cli, requires Node 24+.
  • MCP:https://your-server/mcp，通过服务器自带 OAuth login connects ChatGPT, Claude, or other MCP clients without installing the CLI. Clients that only support local MCP servers can run the CLI's artifact-site mcp. By default, the CLI and MCP create private artifacts and do not generate standalone share links. When a user asks to share the main link, use artifact-site visibility <slug> unlisted or MCP's artifact_site_set_visibility. Example instruction for coding agents: use the /for-agents.md agent guide, CLI, or MCP server; "publish this and give me a link" is one instruction, and sites can also be updated, searched, and read in the same way.

Team deployment and configuration Start from .env.example and follow SELFHOST.md for production deployment. ARTIFACT_PUBLIC_URL must use a stable public address; login callbacks, request origin checks, and the address provided to agents are all derived from it. You can configure a reverse proxy, or set ARTIFACT_WITH_CADDY=on and ARTIFACT_DOMAIN to enable Caddy and automatic certificates. Publishing policies are optional: login requires OIDC; token provides a Bearer token for scripts or agents; open lets anyone who can access the service publish, suitable for trusted internal networks. Anonymous publishing supports separate quotas and expiration times. You can connect Google or OIDC providers such as Keycloak, Logto, Authentik, Okta, and Auth0. Accounts own sites and folders; agents can obtain long-term tokens through device login. By default, ARTIFACT_ENFORCE_OWNERSHIP=on enables account-based access control after OIDC is configured. Add an administrator's verified login email to ARTIFACT_ADMIN_EMAILS to enable the admin console. Set ARTIFACT_WITH_GOTENBERG=on to enable Office previews. The admin console can manage takedowns, quotas, anonymous site expiration, and policy toggles, choose the default visibility, and schedule backups. The application is released as a single Docker image, with Postgres built in for single-host setups. Use the prebuilt imageYou can set ARTIFACT_IMAGE and run make pull. make doctor checks the configuration; make backup and make restore handle backup and restore. For multiple replicas, external Postgres, and S3-compatible storage, see DEPLOY.md.

Technical highlights and limitations

  • Application and storage: Next.js provides the UI and API, Postgres stores metadata, and files are stored on local disk or S3-compatible storage. External databases and object storage support multiple application replicas.
  • Immutable versions: every upload or edit writes a new file version, preserving historical content. Optimistic locking (expected_version) detects concurrent update conflicts.
  • Content isolation: previews use a sandboxed iframe without allow-same-origin and a strict CSP to isolate uploaded content from the platform. Path checks and decompression limits defend against path traversal and ZIP bombs.
  • Hosted pages run in a sandbox and cannot use the platform login session. External API connections require an origin allowlist.
  • The platform hosts artifact files and does not run application backends or build tasks; PDF and Office documents do not support visual HTML editing; scanned images are not automatically OCRed.
  • Cloud agents cannot directly access 127.0.0.1 on the user's computer.
  • The first run downloads dependencies and builds the image, which takes a long time.

Use cases artifact-site is suitable for individuals or teams to centrally manage pages and documents produced by multiple AI agents, when unified sharing, comments, access control, and version management are needed. It is also suitable for scenarios where you want to host these works on your own infrastructure while letting agents continuously update them via CLI or MCP. Local development requires Node 24+ and Docker; for the contribution process, DCO sign-off, and CI requirements, see CONTRIBUTING.md. The project is dual-licensed under the Apache License 2.0 or MIT license.

Related tools

Browse category