AI agent integration
Manage your monitors by talking to Claude, Cursor, or VS Code through the Drumbeats MCP server.
npx -y @drumbeats/mcpnpx -y @drumbeats/mcpAdd that to your MCP client's config and you can run monitoring from a conversation:
"Create a cron monitor for my nightly backup at 02:00 UTC."
"What is my uptime this month?"
"Show open incidents and acknowledge the most recent one."
"Is the SSL certificate for example.com about to expire?""Create a cron monitor for my nightly backup at 02:00 UTC."
"What is my uptime this month?"
"Show open incidents and acknowledge the most recent one."
"Is the SSL certificate for example.com about to expire?"The Connect AI page has ready-to-paste config for Claude Desktop, Claude Code, Cursor, and VS Code.
Connect it#
- Mint an account-scoped key on the Connect AI page. It works across every project you own.
- Choose the scopes. Monitor and diagnostic tools need nothing extra, but creating or updating projects needs
manage_projects. - Install the server with
npx -y @drumbeats/mcpand paste the config into your client. - Ask it something.
Skip the key entirely if you only want the diagnostic tools. Leave DRUMBEATS_API_KEY unset and check_http, check_ssl, and check_dns still work.
What it can reach#
Projects and monitors. List projects the key can see. Create and update projects when the key carries manage_projects. Inside a project, create, list, fetch, update, pause, and resume monitors of any type.
Observation and triage. Pull a monitor's recent pings and response times, read the project-wide uptime rollup, list incidents, and acknowledge or resolve them.
Diagnostics without a key. Ad-hoc HTTP, SSL, and DNS checks, useful for a quick "is this up right now" without touching your monitors.
What it cannot do#
Nothing is deleted. This release exposes no destructive tools. Monitors and incidents can be created, updated, paused, resumed, acknowledged, and resolved, never deleted.
It cannot exceed your key. The server reaches only the projects your key resolves to, and only the operations its permission scopes allow. A key minted without manage_monitors cannot create a monitor no matter how the request is phrased.
Give the agent the narrowest key that does the job. A read-only key is enough for triage and reporting, and it cannot change your alerting while you are asking questions about it.
What happens when it breaks#
| Situation | What you see | What to do |
|---|---|---|
| The key lacks the scope | A 403 with INSUFFICIENT_API_KEY_SCOPE | Mint a key with the scope, or do it in the dashboard |
| The key expired | A 401 | Keys last 1 to 365 days. Mint a new one |
| The agent creates a monitor nothing pings | The monitor sits in its initial state and never alerts | Wire the pings using a language guide |
| The agent creates a monitor with no notification group | Incidents open, nobody is paged | Assign a notification group |
The third and fourth rows are the common ones. A monitor created conversationally is real, and it is inert until your job pings it and a group is attached.
Next#
Monitor types for what the agent can create. Ping API for the wiring it cannot do for you. REST API for the key scopes that bound it. Alternatives if you are still choosing a vendor.