Skip to main content

Tempo MCP Server Examples & Scenarios

These real-world operational scenarios illustrate how AI assistants debug distributed microservice architectures and isolate performance bottlenecks using the TalkOps Tempo MCP Server.

Scenario 1: Latency Spike Root Cause Analysis​

User Prompt:

"The /checkout endpoint has spiked to 4.2 seconds response time over the last 15 minutes. Find the slow traces and identify which service is responsible."

What Happens:

  1. The AI calls tempo_traceql_search:
    • service="frontend"
    • min_duration="3s"
    • start="15m"
  2. It identifies the slowest trace ID (a1b2c3d4e5f60718293a4b5c6d7e8f90).
  3. It calls tempo_summarize_trace(trace_id="a1b2c3d4e5f60718293a4b5c6d7e8f90").
  4. The tool executes critical path analysis and returns:
    • Suspected Root Cause: inventory-service spent 3.85 seconds in span SELECT * FROM inventory FOR UPDATE.
    • Time Gap Disambiguation: Confirms 0ms async lag; the 3.85s delay was synchronous database row locking on the inventory table.
  5. The AI presents the findings and recommends reducing lock granularity or adding an index.

Scenario 2: Pivot from Loki Log Line to Distributed Trace​

User Prompt:

"I found this error in Loki: ERROR 2026-08-20T14:22:01.124Z [order-svc] Payment processing failed trace_id=8f29ab0c47119e8312d4a5b6c7d8e9f0. What happened across all downstream services during this request?"

What Happens:

  1. The AI calls tempo_get_trace_from_log(log_line="ERROR 2026-08-20T14:22:01.124Z [order-svc] Payment processing failed trace_id=8f29ab0c47119e8312d4a5b6c7d8e9f0").
  2. The tool parses the hexadecimal trace ID and retrieves the full multi-service span tree.
  3. The AI reports the failure path:
    • order-svc (HTTP 500) → called payment-gateway (HTTP 502) → called external processor api.stripe.com (Connection reset by peer after 5,000ms timeout).

Scenario 3: Service Dependency Graph & Blast Radius Analysis​

User Prompt:

"Show me upstream and downstream dependencies for payment-service and their error rates."

What Happens:

  1. The AI calls tempo_get_service_dependencies(service="payment-service").
  2. The tool parses live metrics-generator data and returns:
    • Upstream Callers: web-frontend (140 req/s, 0.2% error rate), mobile-gateway (85 req/s, 0.1% error rate).
    • Downstream Targets: fraud-detection (225 req/s, 0.0% error rate), stripe-adapter (225 req/s, 12.4% error rate).
  3. The AI flags stripe-adapter as the active point of failure impacting checkout transactions.