Open-source, self-hosted MCP server. Private until you share. Revocable links. MIT.
This page was written and rendered by ArtifactSmith itself, in its bold style, and published as a share link.
ArtifactSmith gives your own agents a real page you can look at, share, and pull back later, on your own server.
Four steps take an agent request from private build to revocable share link.
The agent calls create with the user's request (verbatim_request) and its notes (source_content or source_files). ArtifactSmith writes and renders the document. The build is private. For a finished page, publish stores it with no model call.
edit makes a new version from the latest one. Every artifact is versioned.
share creates a public /s/ link for one exact version.
unshare makes that link return 404, any time.
Workspaces: tokens belong to one workspace and carry a permission set (create, read, edit, publish, export, share, delete). Knowing an artifact id from another workspace does not grant access. Tokens are stored as SHA-256 hashes.
Each card links to a live share link.
Dark navy-to-violet hero, huge white headline, saturated cards.
Open live demo →Magazine feature, serif headline, thin rules, one deep red accent.
Open live demo →Dark mode, monospace, neon green and cyan, sections as terminal windows.
Open live demo →White grid, huge black grotesque type, signal orange on numbers and rules.
Open live demo →Cream page, candy-colored rounded cards, thick outlines, offset shadows.
Open live demo →The default quiet page; white, one accent, boldness in one place.
Open live demo →Downloads open offline and do not phone home.
HTML is a sanitized, self-contained page. PDF uses WeasyPrint. DOCX uses python-docx. XLSX uses openpyxl.
Set up the server, then point your MCP client at the API and send the token as a bearer header.
To use the copy I run instead, in Claude.ai: Customize → Connectors → + Add → Add custom connector. Paste https://artifactsmith.neuralami.ai/mcp, select Connect, and sign in with your email. The connector then works in the Claude desktop and phone apps too. Free plans allow one custom connector. On Team and Enterprise plans an owner adds it first.
ChatGPT Free, Go, and Plus cannot add a custom MCP server. Pro can add one with read and fetch access only, so it cannot build pages. A ChatGPT Business, Enterprise, or Edu admin can add it on the web; it signs in with OAuth.
cp .env.example .env # Set OPENAI_API_KEY (or AM_LLM_KEY). ./scripts/ensure-local-env.sh docker compose up -d --build docker compose exec server artifactsmith token add --name agent --workspace alpha
Then point your MCP client at {AM_API_URL}/mcp (API port 8780) and send the token as a bearer header:
URL: {AM_API_URL}/mcp Header: Authorization: Bearer <your-token>
Works with any OpenAI-compatible model endpoint (default model gpt-4o-mini). Published image: ghcr.io/vrijsinghani/artifactsmith:0.1.0 (linux/amd64).
Scripts are stripped (CSP script-src 'none').
There are no comments or collaboration.
You self-host it. To see it work before installing, I run a copy at artifactsmith.neuralami.ai/try. The share links on this page are the demo.
You bring your own model key.
Most publish HTML or Markdown the agent already wrote. ArtifactSmith takes the request plus the agent's sources and does the writing and rendering itself, adds named styles, PDF/DOCX/XLSX output and per-workspace tokens.
Yes. Builds stay private until you call share, and unshare kills the link. The process sends no usage data anywhere.
Claude.ai: Customize → Connectors → + Add → Add custom connector. Paste https://artifactsmith.neuralami.ai/mcp, select Connect, and sign in with your email. The connector then works in the Claude desktop and phone apps too. Free plans allow one custom connector. On Team and Enterprise plans an owner adds it first.
github.com/vrijsinghani/Artifactsmith is the copy you run yourself, where Claude Desktop, Claude Code, and Cursor use a token. On this hosted copy they sign in with OAuth at https://artifactsmith.neuralami.ai/mcp, with a pasted token as the alternative.
ChatGPT Free, Go, and Plus cannot add a custom MCP server. Pro can add one with read and fetch access only, so it cannot build pages. A ChatGPT Business, Enterprise, or Edu admin can add it on the web; it signs in with OAuth.
ChatGPT users can use it from OpenAI's Codex CLI instead: run codex mcp add artifactsmith --url https://artifactsmith.neuralami.ai/mcp, then codex mcp login artifactsmith, and sign in with your email.
It's free and MIT-licensed. You pay only for your own model calls and hosting.
Preview pages run on their own cookieless origin; private-network links and likely secrets fail the build. It went through an independent audit (Semgrep, Bandit, gitleaks, pip-audit, Trivy) and a written threat model.