What can I get my agent to publish to a link?
Seven recipes, each ending in a prompt, a command or steps you can follow. The weekly client report on one self-updating URL, the same report kept current by a ChatGPT Dot, the dashboard your agent just built, a Claude artifact worth keeping, an agent that reports its own runs, a morning report of who opened what you sent, and client comments sent to the tools your team works in. Every recipe works from a chat, a workflow tool, CI, or a terminal.
No card · 3 pages free · about a minute
What can I get my agent to publish to a link?
Anything it renders as a single HTML file: the weekly client report, the dashboard it just built, a Claude artifact worth keeping, or its own run report. Publish once and the link stays private and unguessable; when the agent re-runs, it updates the same URL instead of minting a new one. Once the link is sent, the agent can also tell you each morning who opened it, and each client comment can go straight to Slack, Linear or GitHub.
The seven recipes
- recipe Publish a weekly client report to one link that updates itself The most common real workflow: one URL the client bookmarks, refreshed in place every week by your agent, an n8n workflow, or CI.
- recipe A ChatGPT Dot that sends your client the weekly report Your Dot gathers the figures, rewrites the report and updates the private link the client already has, then tells you what changed. You hear when they open it.
- recipe Host an HTML dashboard your agent built One self-contained file to one private URL: drag and drop, one tool call, or one curl. Refresh it in place or let it expire on its own.
- recipe Move a Claude artifact to a private link that will not expire Off the public claude.ai URL and onto a private page with a passcode, your own domain, and updates that keep the same link.
- recipe Have your agent publish its own run report as a link End every run with a URL, not a wall of terminal text. One standing status page the agent itself keeps current.
- recipe Have your agent report each morning who opened what you sent A scheduled prompt for Grok Bot, Claude, ChatGPT or any MCP client: who opened each page through their own link, and what they commented. It only reads.
- recipe Send client comments to Slack, Linear or GitHub Paste a Slack or Discord webhook URL and each comment arrives as a message, or file it as an issue through n8n, Zapier or Make.
Why these six
They follow what people actually publish, not what a pricing page wishes they published. Across real Stacktree publishes, recurring client briefs are the single biggest category, reports and audits sit right behind them, and dashboards follow. Proposals and slide decks, the things delivery tools usually lead with, are around one percent, so they are not here.
The first five are the same primitive wearing different clothes: a single self-contained HTML file goes in, a private unguessable URL comes out, and the next run replaces the content in place so the link you already sent keeps working. Gates layer on top when the audience is a client: a passcode on any plan, a company-email check from Solo, your own domain from $19 a month.
The last two come after the send. A scheduled prompt reads which pages were opened, by whom through their own links, and what comments came in, then reports in your chat without changing anything. And each comment a client leaves can go straight to a Slack or Discord channel, or become a Linear or GitHub issue, through a webhook.
The publishing recipes end with three doors
You should not have to read a setup guide to try any of this. Each publishing recipe closes with three executable exits, in increasing order of commitment. The Dot recipe and the read report close with a prompt instead, and the comment hand-off with setup steps, because all three work on pages in your account.
- One line pasted into any agent.
Fetch and follow https://stacktr.ee/prompt.md, plus what to publish. Works in anything that can fetch a URL. - A one-time install.
npx stacktree-installfor CLI agents, or the Claude.ai connector athttps://api.stacktr.ee/mcp. - A curl with no signup.
POST https://api.stacktr.ee/sitespublishes anonymously; the link lives 24 hours and can be claimed later.
To see what the output looks like before you run anything, the examples gallery is a set of complete, fictional deliverables served as live Stacktree pages.
Frequent questions
What can an AI agent publish to Stacktree? +
publish_html over MCP and gets back a private, unguessable URL; update_site replaces the content in place so the same link stays current.Do these workflows need an account? +
Which agents and tools do these recipes work with? +
Does the person opening the link need an account? +
Related guides
- Claude Code wants to publish an artifact The permission prompt says the tool uploads the file to claude.ai.
- Example deliverables Live pages showing the finished form: reports, proposals, audits.
- Share an HTML file with a client The handoff itself: publish, gate, send one link, revoke if needed.
- Publish from n8n The community node that gives any workflow a publish step.
- Publish from Grok Bot The zero-install path: the bot browses to Stacktree and publishes.
- All use cases The wider catalogue: notebooks, Storybook, diagrams, internal tools.
- Every agent integration Claude Code, Cursor, Codex, Claude.ai, and any MCP client.
- How to host a dashboard your AI agent built An AI agent builds a dashboard as HTML, but HTML is not a link.
Pick a recipe and run the last block on the page.
The first publish is anonymous and needs no account. It lives 24 hours, long enough to decide.
Sign up free →