The watch, over MCP
Your AI client can read and run your watch directly: list sites, check incidents, add a site, tune how often alarms re-fire. Same account, same caps, same rules as the dashboard.
1. Create a key
In Account, create an MCP API key. It is shown once and stored hashed. Revoke it there any time; revocation is immediate.
2. Connect your client
Claude Code, one line:
claude mcp add --transport http promarkseo \ https://promarkseo.com/api/mcp \ --header "Authorization: Bearer pmk_YOUR_KEY"
Any other MCP client, the JSON form:
{
"mcpServers": {
"promarkseo": {
"type": "http",
"url": "https://promarkseo.com/api/mcp",
"headers": { "Authorization": "Bearer pmk_YOUR_KEY" }
}
}
}The endpoint speaks Streamable HTTP (JSON-RPC over POST), tools only. Every request needs the Bearer key.
3. The tools
list_sitesEvery watched site with its current status and last check time.site_statusOne site in detail: baseline tags, open incidents, re-alarm cadence.add_sitePut a site on the watch. Plan caps and subscription state apply.list_incidentsIncidents across all sites, filter by open or resolved.set_realarmHow often open incidents sound again, account-wide or per site. 0 is off.
The floors
set_realarm takes seconds; while an incident stays open, every wired channel sounds again on that cadence until it is fixed. Email is strictly capped at one re-fire per hour to prevent spam. Chat webhooks (Slack, Discord, your own endpoint) may go as low as every 5 seconds. 0 turns re-fires off; the opening alarm always sounds.
Scope and safety
A key sees exactly one account: its sites, its incidents, nothing else. Keys never grant login, password, or billing access. We store only the key’s hash; if you lose one, revoke it and create another.