Can I create and manage agent alerts from my coding agent?

Yes, Rius lets your coding agent list, create, update and delete alert rules and notification channels, and acknowledge firings, through the Rius MCP server. Connect with your Rius account and the tools act with your role. Admins manage alerting, and every member can read alerts and acknowledge a firing. A headless agent can manage rules and channels with an API key, but it never acknowledges a firing.

How it works

The Rius MCP server exposes alert tools your client calls when you ask in plain language. They fall into three groups:

• Read: list_alerts, get_alert, list_alert_rules and describe_alert_catalog.
• Manage rules and channels: create_alert_rule, update_alert_rule, delete_alert_rule, create_alert_channel, update_alert_channel and delete_alert_channel.
• Acknowledge: ack_alert marks a firing as seen, with your name and time, and stops further notifications for it. The firing still resolves on its own once the condition stops matching.

Every tool carries hints that say whether it only reads or changes something, and whether a change is destructive, such as an update that replaces a whole list. Clients that honor the hints can auto-approve reads and ask you to confirm writes.

What a tool can do depends on how you sign in:

• OAuth session: this is your Rius account. Reading and acknowledging are open to every member. The alert-management tools work only for admins, as in the console.
• Agent API key: the key reaches the one workspace it was created in. A bare key is read-only. A key minted with alerts:manage can manage alert rules, and one with channels:manage can manage distribution channels. No key can acknowledge a firing, so ack_alert is always refused.

The Rius plugin for Claude Code, Codex (beta) or Cursor (beta) already registers this server as rius, so you do not add it again.

Two behaviors to know. Saving a change to an alert resolves its open firings and evaluates the rule afresh, so a condition that still holds opens a new firing. And the errored-span threshold of the Swallowed Error alert is set only through the MCP server.

Set it up

1. Connect your client. In Claude Code, run this once, and the OAuth sign-in opens the first time a session uses the server:

   claude mcp add --transport http rius https://mcp.eu.console.rius-glassflow.com/mcp

   Codex and other clients have their own connect snippets on the MCP page.

2. For an agent that runs without a browser, such as a CI job, have an admin open the console’s MCP page and choose Connect an agent. Tick Manage alert rules (alerts:manage), Manage notification channels (channels:manage), or both. Only this dialog mints these two scopes, and the Create API key dialog in Settings does not offer them. The key is shown once, so store it as a secret. Then send it as a bearer header. If the Rius plugin is installed, register the key under its own name, such as rius-key:

   claude mcp add --transport http rius-key https://mcp.eu.console.rius-glassflow.com/mcp \
     --header "Authorization: Bearer ${RIUS_API_KEY}"

   Without a manage scope, the alert-management tools are still listed, but each one replies that the key is not allowed and changes nothing.

3. Check the connection by asking “Who am I in Rius, and which workspaces can I access?” The client calls get_me and answers with your workspaces. If you belong to several organizations and a workspace name is ambiguous, pass it as org/workspace.

4. Ask for what you need in plain language:

   • “List the alerts that fired in the last day.”
   • “Create a webhook channel and attach it to a new alert on error rate.”
   • “Acknowledge the open Tool Loop firing.”

   Your client picks the tool, and a good habit is to have it call describe_alert_catalog before it creates a rule.

What you get

• Alerting from the editor: list firings, inspect one alert, and review rules and channels without opening the console.
• Rules and channels as conversation: create, change and remove custom rules and webhook, email or Slack channels from your coding agent session.
• Acknowledgment where you work: as a signed-in member, you acknowledge a firing from the same session where you investigate it.
• Safe automation: a scoped key lets a scheduled agent maintain alert rules in one workspace. It cannot mint other keys, read spend notices or acknowledge firings.
• A record you control: Rius logs each MCP call with its tool, outcome and structural arguments. Free text such as search terms, webhook URLs and email recipients is never kept, and tool results are never stored. Call-log rows are kept for 30 days.

The Alerts page covers firings, renotify, rule health and root-cause analysis in full.

When something else fits

• Use the console for anything the MCP server does not offer. Nothing it offers can create, rename or delete workspaces or change members, and key management for a headless agent also stays in the console.
• If you want to look through firings with filters and analysis status, the Firings page in the Alerts section is a better view than a chat transcript.

Related