AI assistants (MCP)

Overview

OKStatus runs a remote MCP server (Model Context Protocol), hosted in Europe like the rest of the platform. Add it as a connector in your AI assistant and you can ask, in a conversation, what is down, how a monitor performed last month or which maintenance windows are planned.

Reading works on every plan, with the rights of a viewer: the assistant sees what a viewer of your organization sees, and nothing more. On Max and Enterprise, the assistant can also make changes — create a monitor, open an incident, schedule a maintenance.

Connector URL

https://mcp.okstatus.eu/mcp

Set it up

Copy the URL above, including the final /mcp — https://mcp.okstatus.eu alone does not work.
Add it as a connector in your assistant.
- Mistral Vibe (formerly Le Chat): add a custom connector, paste the URL in the Server field and leave authentication on OAuth 2.1. The advanced OAuth settings stay off — no client ID is needed.
- Claude: Settings → Connectors → Add custom connector, then paste the URL.
- Claude Code:
  Terminal window
```
claude mcp add --transport http okstatus https://mcp.okstatus.eu/mcp
```
  Without --transport http, Claude Code takes the URL for a local command and the connection fails.
- VS Code (GitHub Copilot): run MCP: Add Server… from the Command Palette, choose HTTP and paste the URL. Or add it to .vscode/mcp.json:
```
{ "servers": { "okstatus": { "type": "http", "url": "https://mcp.okstatus.eu/mcp" } } }
```
- Another assistant: assistants are enabled one by one, and the OKStatus sign-in refuses one it does not know. Tell us which one you use at contact@okstatus.eu.
Sign in to OKStatus. The assistant opens the OKStatus sign-in page. Log in with your usual account (password, SSO or social sign-in), then accept the access it asks for.

Your account must belong to an organization. If you belong to several, the assistant can list them and query the one you name; otherwise it uses your default organization.

What you can ask

The assistant picks the right tool by itself. Today the server offers:

Tool	What it returns
`list_organizations`	The organizations you belong to, with their plan and your role
`get_status_overview`	How many monitors are up, down or paused, and the open incidents
`list_monitors`	Your monitors, filterable by status, type, name or group
`list_monitor_groups`	Your monitor groups (the folders of the Monitors page) and their monitors
`get_monitor`	One monitor’s settings, 24-hour uptime and its 10 latest checks; for an AI model monitor, its TTFT, TPOT and ITL over 24 hours
`get_monitor_uptime`	Uptime and minutes of downtime over 24 h, 7, 30 or 90 days
`list_monitor_checks`	The latest checks of a monitor, per region
`list_incidents` / `get_incident`	Incidents and their timeline
`list_maintenances`	Scheduled, ongoing and past maintenance windows
`list_status_pages`	Your status pages: address, whether each is published or password-protected, and the monitors it shows
`list_alert_channels`	Your alert channels (name, type, health, attached monitors) — never their webhook or token
`list_private_agents`	Your private agents: status, version, host, load and the monitors they run
`get_uptime_chart`	A picture of the daily availability of monitors or a group over 7, 30 or 90 days

Examples:

“Is anything down right now?”
“What was the uptime of the payment API last month?”
“Which maintenance windows are planned this week?”
“Show me a chart of the Payments group over the last 30 days.”
“Which monitors have no alert channel?”
“Are my private agents up to date?”

Making changes (Max and Enterprise)

On the Max and Enterprise plans, the assistant can also act:

Tool	What it does
`create_monitor`	Creates a monitor
`update_monitor`	Changes a monitor’s settings (target, interval, regions, tags, accepted codes, keyword, port, DNS), moves it into a group, pauses or resumes it
`set_monitor_alert_channels`	Attaches existing alert channels to a monitor
`get_sla_report_link`	Generates an SLA report (PDF) and returns a private download link, valid one hour
`send_sla_report`	Generates an SLA report (PDF) and e-mails it once to up to five recipients
`create_incident`	Opens an incident
`add_incident_update`	Adds an update to an incident, including its resolution
`create_maintenance`	Schedules a maintenance window
`cancel_maintenance`	Cancels a scheduled maintenance window

Examples:

“Create an HTTP monitor for https://api.example.com/health, every minute, from Paris and Frankfurt.”
“Move the checkout monitors into the Payments group and attach the on-call channel to them.”
“Pause the staging monitors.”
“E-mail last month’s SLA report for the Payments group to finance@example.com.”
One consent. When you add the connector, the OKStatus consent screen asks for reading (okstatus:read) and for creating and updating (okstatus:write) together. If you connected before October 2026, your connection only reads: remove the connector and add it again.
Your role still applies. A viewer cannot change anything through the assistant either.
Same rules as the dashboard. Monitor limits, minimum check interval and notifications are those of your plan: a change made through the assistant is the same change as the one made by hand.
Nothing is deleted. No tool removes a monitor, an incident or a status page.
No secrets through the assistant. Alert channels are created in the dashboard, because a channel holds a webhook URL or a token; the assistant can only attach channels that already exist.
Reports count against your monthly quota, exactly as those generated from the Reports page.
Traced. The audit log names you and the assistant used.

Security and limits

Read-only below Max. On the Free and Pro plans, and for viewers on any plan, no tool changes your monitors, incidents or settings, whatever was accepted on the consent screen.
Where your data goes. What the assistant reads is sent to the assistant you chose. We recommend Mistral Vibe, hosted in the EU; with another assistant the consent screen warns you that data may leave the EU.
No secrets. Credentials embedded in a monitor’s URL (https://user:pass@host) are removed before anything reaches the assistant.
Rate limit. 300 requests per 5 minutes per user, of which at most 30 changes, separate from your organization’s API quota.
Revoking access. Remove the connector in your assistant; it then has to sign in again to read anything.

Assistant, API or Terraform?

The three reach the same monitors, groups and alert channels, with the same plan limits.

	Best for	Can delete	Handles secrets
AI assistant (MCP)	Questions, and one-off changes in a conversation	No	No — alert channels and credentials stay in the dashboard
REST API	Your own scripts and CI pipelines	Yes	Yes
Terraform	Monitoring kept in version control, next to the infrastructure it watches	Yes	Yes