@perfana/mcp

MCP server that exposes Perfana test run data to Claude Desktop (or any MCP-compatible client).

Tools

| Tool | Description | |---|---| | get_test_run | Metadata, status, and configuration for a test run | | get_transaction_stats | Response times (avg/p95/p99), throughput, error rates, Apdex scores | | get_recent_runs | Recent runs for a SUT/environment/workload — for trend comparison | | compare_runs | Side-by-side regression diff between two runs | | get_test_run_configs | Configuration items captured during a run (JVM flags, pool sizes, feature flags, etc.) | | get_config_diff | Diff configuration items between two runs — highlights added/removed/changed keys | | get_check_results | SLO / requirements check results — which objectives passed or failed | | get_adapt_results | Adapt regression analysis: overall verdict, classified regressions, causal chains, hypotheses | | get_deep_links | Resolved dashboard/tool links associated with a test run | | get_performance_rankings | Top N transactions ranked by slowest, highest throughput, highest impact, or highest error rate | | get_error_analysis | Error summary, errors grouped by HTTP status code and by transaction | | get_error_details | Detailed error instances for a specific transaction — request/response headers and body | | get_available_metrics | List all dashboards, panels, and metric names available for a test run | | get_metric_trends | Time-series data for a specific metric during a test run | | list_connected_sources | Discover connected data sources: Grafana, Tempo, Pyroscope, Dynatrace | | get_grafana_dashboard_snapshot | Min/max/avg/last summary for all panels in a Grafana dashboard | | get_slow_traces | Slowest distributed traces from Tempo/Jaeger, with optional service/scenario/transaction filter | | get_trace_detail | Full span breakdown for a specific trace | | get_error_traces | Distributed traces containing errors from Tempo/Jaeger | | get_flamegraph | CPU flamegraph data from Pyroscope in collapsed-stack format | | get_hotspots | Top N hottest methods from Pyroscope, sorted by CPU sample count | | get_dynatrace_problems | Dynatrace-detected problems during the test run time window |

Setup

1. Configure Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "perfana": {
      "command": "npx",
      "args": ["-y", "@perfana/mcp"],
      "env": {
        "PERFANA_API_URL": "http://localhost:3001/api",
        "PERFANA_API_KEY": "YOUR_PERFANA_API_KEY"
      }
    }
  }
}

Generate an API key in Perfana under Settings → API Keys.

3. (Optional) Add Obsidian MCP

To write analysis notes directly to your vault, also add the obsidian-mcp server.

Usage examples

Once configured, just ask Claude Desktop:

"Analyze test run perfana-gatling-myapp-1741859234 and write a performance report to my Obsidian vault under Performance Tests/2026-03-16 myapp peak-load.md"

"Compare the last 3 test runs for myapp in the production environment and tell me if there are any regressions"

"What was the Apdex score for /api/checkout in test run X?"

"Show me the top 10 slowest transactions in test run X and which ones are regressions vs baseline"

"Are there any CPU hotspots in the afterburner service during test run X?"

Investigation workflow

For root cause analysis, a typical agent workflow is:

1. get_adapt_results — get overall verdict and classified regressions
2. list_connected_sources — discover which investigation tools are available
3. get_performance_rankings (slowest/highest_impact) — identify bottleneck transactions
4. get_slow_traces / get_error_traces — drill into distributed traces
5. get_flamegraph or get_hotspots — find CPU hot methods (if Pyroscope connected)
6. get_grafana_dashboard_snapshot — check container/JVM metrics

Environment variables

| Variable | Default | Description | |---|---|---| | PERFANA_API_URL | http://localhost:3001/api | Base URL of the Perfana API | | PERFANA_API_KEY | (required) | Perfana API key — sent as Authorization: Bearer <key>. Generate one in Perfana under Settings → API Keys. |