MXCP Observability

MXCP Observability is a monitoring platform for MCP server deployments that provides traces, metrics, and audit logs through a unified API. It connects to multiple MCP instances, aggregates telemetry data, and gives you fleet-level visibility into what your AI agents are actually doing in production. Think of it as the control plane for all your MCP servers.

Standard APM tools don't understand MCP semantics. Without MXCP Observability, you can't see which tools are being called, how often, by whom, or how long they're taking. When an AI workflow slows down, you're guessing whether the bottleneck is the model, the MCP server, or the data source. And in regulated environments, you need audit trails that prove compliance—ad-hoc logging doesn't cut it.

We calculate P50, P95, and P99 latencies per endpoint so you know typical and worst-case performance. We track request counts, error rates, and trends over configurable time windows. You can see which tools and resources are most used, which are failing, and how performance changes over time. All metrics are available via API for integration with your existing dashboards.

MXCP Observability receives OpenTelemetry traces via the standard OTLP protocol on port 4318. You can visualize the complete request flow with span hierarchies, timing breakdowns, and status codes. When a request touches multiple services—your MCP server, a database, an external API—you see exactly where time is spent and where errors occur.

Audit logs are queryable by user, operation, time range, policy decision, or trace ID. You can investigate specific incidents, generate compliance reports, or analyze access patterns. The query interface supports both the REST API and CLI, so you can integrate audit queries into your existing workflows or automation pipelines.

The simplest deployment is a single Docker container that runs the mxcpd daemon and optional web dashboard. For production, we provide Kubernetes Helm charts with sidecar or daemonset patterns for multi-instance monitoring. You can also install just the mxcpctl CLI via pip for headless environments or scripted access. All components are stateless and horizontally scalable.

Yes. Since MXCP Observability uses OpenTelemetry, you can export traces to any compatible backend—Grafana, Datadog, New Relic, Jaeger, or your existing observability stack. The REST API makes it easy to pull metrics into custom dashboards or alerting systems. We don't lock you into our visualization; use whatever tools your team already knows.

MXCP Observability holds traces and metrics in memory with configurable retention (default: 24 hours for traces, 7 days for metrics). For long-term storage and compliance archival, export data to your existing data warehouse or observability platform. This keeps the daemon lightweight and lets you use your preferred storage infrastructure.