Ask your Grok Bot to publish it. That is the whole setup.

Why there is nothing to install

Most integrations on this site need something: a node in n8n, an MCP server in Claude Code, a connector in Claude.ai. Grok Bot needs nothing for a first page, because it can sign into tools and navigate them the way a person does, in a browser on its own cloud computer.

That means the thing standing between an agent and a service is not whether an integration exists. It is whether the first useful action can be done without an account. Stacktree's first publish is anonymous by design, so the bot can complete the job in one pass.

Connect your account in one message

Grok Bot can now add custom MCP connectors when you ask it to, and Stacktree's MCP server is one. Connected, the bot stops driving a browser and calls Stacktree directly, with everything an account has: pages that do not expire, a passcode or an email check, a link per person that tells you when they opened it, comments from readers, and updates to a page you sent last week. Ask any of your bots:

Add Stacktree as a custom MCP connector. It is a remote HTTPS server
that signs in with OAuth, so leave the HTTP headers empty.
Server URL: https://api.stacktr.ee/mcp

The bot shows an approval card with the URL, "OAuth sign-in" and "no HTTP headers" before it saves anything. Approve it, press Authorize on the Stacktree card that follows, and Stacktree's sign-in page opens in your browser listing what the bot would be able to do. Click Allow and the card reads "29 tools, Added". The connector belongs to your Grok account, so every bot on it can use Stacktree from then on. No API key is involved, and you can take the access back at any time from your Stacktree API keys page.

We ran this on 2 October 2026. After sign-in we asked the bot for a private test page with a link for one named person. It called publish_html, then create_share_link, and came back with the page, the named link and the correct expiry date. One thing caught us out: the desktop app finishes sign-in on port 8787 of your own computer, so if something else is using that port (Cloudflare's wrangler dev does by default), sign-in ends on a "not found" page. Stop that program and reload the page.

What we actually tested

On 20 August 2026 we gave a Grok Bot one instruction: "Create a standard intro page and publish it to stacktree." No plugin, no key, no integration. It reported that it would look up how Stacktree pages work, found the house design guide, wrote a one-pager to that specification and published it, then returned the live URL and described the result correctly without being asked: unlisted, viewers need no account, expiring the next day. It also offered the right next step, which was claiming the page so it could update that URL in place instead of minting a new one.

Our server records match: no account attached, no payment, one file, 24 hour expiry. You can read the page it produced, which is still up because it was claimed into an account afterwards. The full write-up is in Grok Bot explained.

Making it repeatable

Grok Bot is built around routines: it learns a workflow from one demonstration and reuses it. That is where publishing in place matters. Once a page belongs to an account, every later run updates the same link rather than producing a new one, so the URL you gave a client on Monday still shows Friday's version. A routine that mints a fresh link every morning is not something you can send anyone.

To get there, connect Stacktree as above and tell the bot to update the page it already published rather than publishing a new one. Without a connection, publish once anonymously, claim the page, and give the bot that page's address for later runs.

Or skip the setup: grab the template

Grok Bot templates (added August 2026) let one person's Bot configuration install into yours in a tap. We publish one that carries all of the above: publish on request, always report the link, its expiry and the claim link, update in place on revisions, and never publish secrets without asking.

Publish work as a private link: add it to your Grok Bot →

Since version 4 (2 October 2026) the template uses the Stacktree connector when your Grok account has it, offers once to add it when it does not, and otherwise falls back to a 24-hour page that needs no account. Connected, pages belong to your account, so routines keep updating the same URL.

When the page should not be open to anyone with the link

The default is an unguessable URL that is never listed or indexed, which is enough for most internal work. When the page carries client material, paid plans add a passcode, or a gate that restricts viewers to a company email domain and lets them through by magic link without creating an account. Expiry and burn-after-read are available on every plan.

Other agents

If you also drive Claude Code, Codex CLI, Cursor or Claude.ai, those speak MCP and take a faster path than browser automation: they call publish_html directly and get a URL back. One command sets all of them up:

npx stacktree-install

See every agent integration for the full list.