An HTML page
An agent can read it, expensively and unreliably: a design is not data.
AI agents
Status pages for AI agents
Agents read your status page and can run it: an MCP server, a Markdown version, and a log that shows what the assistant did.
Why this already matters
More and more incident work does not happen in a browser:
- Whoever is on call asks an assistant what is failing instead of opening a dashboard
- An agent has to publish a report while the human is still reading logs
- A customer's system wants to read your status as text, not HTML
Where agents get stuck
An API with no key scopes
Either the agent gets full access or none. Neither is the right answer.
Tools that leave no trail
When an assistant publishes something, there has to be a record of who did it and when.
What we ship for agents
An MCP server
api.statusas.lt/mcp connects Claude, ChatGPT, Cursor and any Model Context Protocol client to your workspace.
Publishing always asks
Every publishing tool requires an explicit decision about notifying subscribers. A model cannot quietly mail a thousand people.
Markdown instead of HTML
Append .md, or send Accept: text/markdown, and the page comes back as text an agent can read cheaply.
Everything in the log
Every action lands in the audit log, marked as coming from an MCP client.
Connecting it
- 1
Create an API key with the access level you want
- 2
Connect the MCP server to your client — OAuth or a key in a header
- 3
Pick the workspace and approve the access
- 4
Let the agent read monitors and write status reports
- 5
Review the audit log — everything it did is there
The dashboard also shows which apps have access, and you can revoke it at any time.
Audit log
- 14:31incident.publish · api-eumcp:claude-desktop
- 14:31subscribers.notify · 1,204approved by j.kazlauskas
- 14:28monitor.read · list_monitorsmcp:claude-desktop
- 13:02apikey.create · scope: writej.kazlauskas
- 12:44page.read · status.acme.lt.mdmcp:cursor
Frequently asked questions
Can an agent accidentally notify our customers?
No. Publishing tools require an explicit decision about notifying subscribers, and every action stays in the log.
Which clients are supported?
Any Model Context Protocol client: Claude Desktop and Claude Code, ChatGPT, Cursor, opencode and others.
Can I grant read-only access?
Yes. An API key carries a scope, and the scope decides which tools the server exposes at all.
Start monitoring in five minutes
One URL, one email address — and you will hear about the next outage from us, not from a customer.