Honeycomb MCP

@Honeycomb show me the P95 latency for the checkout service over the last hour, broken down by region

Based on trace data from the last hour, P95 latency for checkout service: US-East 245ms, US-West 312ms, EU-Central 198ms. EU-Central shows the fastest response times. US-West experienced a spike at 14:23 UTC reaching 890ms for approximately 3 minutes, likely correlating with the deployment event logged in your timeline. Overall request volume was consistent across regions at roughly 1,200 requests per minute.

This example demonstrates querying observability data across dimensions. Requires OAuth scope for read access to datasets. Honeycomb's query engine aggregates high-cardinality trace data, so response time depends on dataset size and time range—wider windows may take 5-10 seconds to return.

@Honeycomb create a trigger that alerts #incidents when error rate exceeds 5% for the payment-api service in production

Trigger created successfully: 'payment-api-error-rate-alert' now monitors the production environment. Alert fires when error rate crosses 5% threshold over a 5-minute window, evaluated every minute. Notifications will post to your configured Slack #incidents channel. Trigger ID: trg_8k3n9x2p. You can view or edit this trigger in the Honeycomb UI under Triggers > payment-api-error-rate-alert.

This write operation requires OAuth scope for trigger management. The MCP creates the alert definition but does not configure the Slack webhook—that integration must already exist in your Honeycomb account settings. Changes take effect immediately and begin evaluating on the next minute boundary.

@Honeycomb analyze the traces from the last deployment of user-service and summarize what changed in terms of database query patterns

Comparing traces before and after the 16:40 UTC deployment: The new version introduced 3 additional database queries per user profile load, increasing average query count from 4 to 7. Most notable: a new SELECT on the preferences table now runs on every request instead of being cached. This added approximately 18ms to P50 latency. However, the deployment eliminated a previous N+1 query pattern on the friends table, which previously caused occasional 2-second spikes. Net effect: more consistent performance with slightly higher baseline latency.

This synthesis example pairs Honeycomb's trace data with AI reasoning to surface deployment impact. Accuracy depends on the deployment marker existing in your dataset and sufficient trace volume for statistical comparison. The AI interprets patterns but cannot access your application code—findings reflect observable behavior only.

Surface latency trends in your team sync without opening dashboards

A backend team runs a 15-minute Friday standup to review the week's performance metrics. The Honeycomb MCP lets the tech lead pull p95 latency and error-rate summaries for each service directly into the Switchy meeting notes. The team discusses trends in the same doc where they assign follow-up tickets, so the context stays linked. This beats opening five Honeycomb dashboards and narrating them over Zoom. The limitation: if you need to compare this week to the same week last quarter, or overlay multiple services on one chart, you'll still use Honeycomb's UI. The MCP shines for quick spot-checks and week-over-week comparisons where the query shape is predictable. Best for teams that already have a standing performance-review ritual and want to reduce the prep overhead.