Observability & Monitoring
fme-mcp integrates enterprise observability out-of-the-box, supporting OpenTelemetry standards, Prometheus metrics, structured JSON logs, and immutable audit trails.
Metrics & Health Probes
When running in Streamable HTTP mode, the server exposes dedicated monitoring endpoints:
GET /healthz: Liveness probe verifying the event loop is active.GET /readyz: Readiness probe testing FME Flow API or local FME Form engine connectivity.GET /metrics: Standard Prometheus-formatted scrape target.
Prometheus Metrics Exposed
| Metric Name | Type | Description |
|---|---|---|
fme_mcp_requests_total |
Counter | Total JSON-RPC tool calls by tool name and status |
fme_mcp_request_duration_seconds |
Histogram | Latency distribution of tool execution |
fme_mcp_active_jobs |
Gauge | Number of concurrent active FME workspace runs |
fme_mcp_rate_limit_rejections_total |
Counter | Rejected requests due to token bucket depletion |
fme_mcp_security_violations_total |
Counter | SSRF, path traversal, or command injection blocks |
Structured Logging
Logs are formatted in structured JSON conforming to the OpenTelemetry schema. Every log entry includes:
- timestamp: UTC ISO-8601 with microsecond precision.
- level: INFO, WARNING, ERROR, CRITICAL.
- correlation_id: Unique UUID per JSON-RPC request for distributed tracing across agent hops.
- tool_name: Invoked MCP tool identifier.
- user_id / client_id: Authenticated caller identity.
Immutable Security Audit Trail
All data transformations, parameter bindings, and workspace executions are appended to an immutable append-only audit log:
{
"timestamp": "2026-11-06T10:45:00.123456Z",
"event_type": "TOOL_INVOCATION",
"tool": "fme_run_workspace",
"principal": "agent-claude-3-7",
"status": "SUCCESS",
"duration_ms": 1420,
"parameters_hash": "sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
"client_ip": "10.244.1.42"
}
Audit records can be forwarded directly to SIEM solutions such as Splunk, Datadog, ElasticSearch, or Google Cloud Operations Suite.