MCP Usage Guide

This guide covers common workflows for investigating incidents and querying cluster state using PulseStream’s MCP tools.

Investigate an incident

When you receive an alert or notice an issue, use the MCP tools to investigate:

"List open incidents and analyze the most critical one"

PulseStream will:

1. List your incident groups filtered by status and severity
2. Trigger AI root cause analysis on the selected group
3. Return findings including probable cause, affected services, and remediation steps

Ask follow-up questions

After an initial analysis, ask follow-up questions in the same conversation:

"What changed in the last deployment that could have caused this?"

"Show me the error logs from the affected pods"

Proactive investigation

You don’t need an active incident to investigate. Query any namespace or service:

"Investigate the production namespace — are there any unhealthy services?"

"Check the payment-service in production — why is latency increasing?"

Cluster-wide overview

Get a broad view of your cluster health:

"What's the overall health of my cluster? Any services I should worry about?"

Tips

• Be specific — mention service names, namespaces, and time ranges when possible
• Use conversation context — MCP supports multi-turn conversations, so follow-up questions retain context
• Combine with your tools — use MCP alongside kubectl, your APM, and log tools for a complete picture