Your agent works for hours. Watch it from anywhere.
Paste one paragraph into Claude Code, Codex, Cursor or any agent that can publish files, and it keeps a live status page for the task — a public URL you can check from your phone or send to whoever is waiting on the work.
One paragraph, pasted once
Paste it at the start of a long task, or put it in AGENTS.md or CLAUDE.md so every run gets a page. The agent fetches /llms.txt for the details, so the prompt stays short and the recipe stays current.
Keep a live status page for this task so I can follow it from anywhere. Build it as one self-contained HTML file and publish it with harvis.dev: fetch https://harvis.dev/llms.txt and follow the "Live status pages" section. Send me the URL once, then redeploy to that same URL after every meaningful step until the work is done.
A page that answers "how's it going?" before you ask
Every agent designs its own, but the instructions ask for the same parts: the goal, a checklist, what it is doing right now, anything it is stuck on, and when it last checked in. An open tab refreshes itself every minute.
Open a page Claude Code kept for a real taskFor the hours you aren't watching the terminal
Check in from your phone
Leave a four-hour refactor running and glance at a URL from the train, instead of reconnecting to scroll back through a log.
Share progress without sharing a screen
Send the link to a teammate, a client or your manager. They get a page that reads like a report, not a terminal.
Catch a stuck agent early
Blockers and questions sit at the top of the page, so a run that is waiting on you doesn't idle until you happen to come back.
Keep a record of the run
The last deploy is a summary of what changed and what's left. Claim it and it stays as the write-up.
Three things happen, and you only do the first
- 1
Paste the prompt
At the start of a task, or once in your agent's instructions file.
- 2
The agent publishes the page
It writes one HTML file, deploys it with npx harvis, the upload endpoint or the MCP tool, and sends you the live URL and a private claim link.
- 3
It redeploys as it goes
After each meaningful step it rewrites the page and deploys again with the same deploy token, so the address never changes.
Any agent that can run a command or make a request
If it can run npx or curl, or call an MCP tool, it can keep the page. Nothing to install and no API key to hand over — the agent deploys anonymously and gives you the claim link.
- Claude Code
- Codex
- Cursor
- Gemini CLI
- GitHub Copilot
- Cline
- Aider
- Windsurf
Questions
What people ask before they leave an agent running.
Who can see the page?
Anyone with the URL. Addresses are random and not listed anywhere, but they aren't secret either, so the instructions tell the agent never to put keys, credentials, customer data or raw logs on the page. For confidential work, say in the prompt how much detail you're comfortable having in public.
How long does the page stay up?
An unclaimed site is permanently deleted 24 hours after its last deploy and can't be restored. Every update counts as a deploy, so the page stays live while the agent is working and for a day after. Open the claim link and sign in free to keep it. If the agent deploys with your account's API key, it lands in your dashboard and doesn't expire.
Will a new link appear with every update?
No. The first deploy returns a private deploy token, and the agent sends it back with each update, so the same address is replaced in place. Running npx harvis again from the same folder does that on its own.
Does it slow the agent down?
Barely. A deploy is one command and takes a couple of seconds, and the instructions ask for an update after each meaningful step, not after every file the agent touches.
Can I choose what the page looks like?
Yes — it's plain HTML. Ask for your team's colours, a progress bar or a timeline in the prompt, or give the agent a template file to fill in each time.
Start your next long run with a link
Copy the prompt, paste it in, and go do something else.