MCP analytics notifications

MCP analytics can reach you in Slack, Microsoft Teams, Discord, email, or any webhook, in two ways: a recurring written summary, or an instant alert on a single event.

Which one you want depends on the question. Use a recurring report for anything you'd think about periodically, like what agents keep trying to do. Use an instant alert for something you'd want to know about immediately, like one of your tools breaking.

The Notifications tab lists every kind of notification in one table, with its cadence and whether it is on. Expand a row to see what it sends, and the destinations you have set up for it.

Recurring reports

A recurring report is an AI-written summary of your MCP traffic, delivered on a schedule. Two are ready to set up from the Notifications tab:

• What agents are trying to do (weekly) groups and ranks agent goals in their own words, and shows which of them keep ending in failure.
• How your tools held up (weekly) covers calls per tool, error rate, the most common error types, and p95 latency.

To set one up:

1. Open MCP analytics and select the Notifications tab.
2. Select Set up on the row of the report you want.
3. Choose where it should go and how often, then create it.

The report's question is already written for you, and you can edit it before saving.

Recurring reports are built on PostHog's AI subscriptions, which are in early access. If the report shows Turn on early access instead of Set up, select it to turn on AI subscriptions in your feature previews first. Your organization also needs to approve AI data processing in its settings.

Because a report summarizes a window of time, it stays useful on a low-volume server and can't flood a channel.

Instant alerts

An instant alert reaches you as soon as something happens that you'd want to know about right away: a failed tool call, captured as $mcp_tool_call with $mcp_is_error.

To create one, select Set up on the A tool call failed row, then choose a channel. PostHog prefills the message with the tool, your server, the agent's client, and its intent. You can customize it later from the alert's own page.

The alert is deduplicated per failing tool: if one tool fails a hundred times in a row you get one message, not a hundred, and each additional tool that starts failing still gets its own. A tool that is still broken reappears after 30 minutes. You can change this under Trigger options on the alert's page, including turning deduplication off.

Manage alerts and reports

Instant alerts are standard CDP destinations, so you can also manage them from the data pipelines page. To use another destination type, create it manually with the same MCP events and filters. Recurring reports appear alongside your other subscriptions, where you can edit, pause, or delete them.

Slack requires the PostHog Slack integration. Microsoft Teams, Discord, and webhook destinations only need a webhook URL.