Using the MCP server

1. Log in and open API tokens in the dashboard sidebar.
2. Name the token after the agent that will use it (for example claude-code) and create it.
3. Copy the token immediately: it is shown exactly once.

Tokens act as you, on your current team. Revoke one from the same page at any time and the agent loses access instantly.

Claude Code

One command in your terminal:

claude mcp add --transport http undownable https://undownable.com/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

Claude Desktop

Add this to the mcpServers section of your claude_desktop_config.json (Settings, Developer, Edit Config):

"undownable": {
  "command": "npx",
  "args": [
    "mcp-remote", "https://undownable.com/mcp",
    "--header", "Authorization: Bearer YOUR_TOKEN"
  ]
}

Any other MCP client

Point it at https://undownable.com/mcp using the HTTP (streamable) transport and send Authorization: Bearer YOUR_TOKEN with every request. Most clients accept a config shaped like:

{
  "url": "https://undownable.com/mcp",
  "headers": { "Authorization": "Bearer YOUR_TOKEN" }
}

That's the whole setup. Your agent discovers the tools on its own; you just talk to it:

• "Monitor https://example.com every 5 minutes and alert me if it goes down."
• "Is anything down right now? Show me this week's incidents."
• "Add an SSL check for example.com and a TCP check for db.example.com:5432."
• "Create a heartbeat monitor for my nightly backup job and give me the ping URL."
• "Set up a public status page with my production monitors on it."
• "Pause the staging monitors while we deploy, then resume them."

Pass these optional fields to create-monitor or update-monitor. Available on every plan, including Free. Repeats continue while an incident is open and unacknowledged, up to the chosen maximum. Omitted fields stay unchanged on update. Create, update, get, and list responses include both settings.

Incident payloads include acknowledged, acknowledged_at (ISO 8601 or null), acknowledged_by ({name} of the account, or null), acknowledged_via (panel, api, mcp, link, or null), and realert_count. A signed link can acknowledge without an account. Acknowledgement stops repeats; recovery resolves the incident.

acknowledge-incident returns acknowledged, team, and the incident payload. An incident outside the selected team returns an error.

History and pagination

list-incidents accepts cursor from meta.next_cursor and limit (default 20, maximum 100). Results are ordered by started_at descending, then ID descending. meta.open_count and meta.unacknowledged_count count ongoing and ongoing unacknowledged incidents before pagination. get-incident returns the same detail payload as REST, including the team and delivery history.

list-monitors includes 30 days of history on each monitor; get-monitor includes 90 days and a four-hour heatmap on its monitor object, matching REST. History rows contain day, uptime, and level (up at 99.95%, dip at 95%, down below 95%, none without data). Heatmap cells contain start, end, level, checks, and ratio; levels are up-1 through up-4, dip, partial, down, and empty.