turbomcp-telemetry

OpenTelemetry integration and observability for TurboMCP SDK.

Features

• Distributed Tracing: OpenTelemetry traces with MCP-specific span attributes
• Metrics Collection: Request counts, latencies, error rates with Prometheus export
• Structured Logging: JSON-formatted logs correlated with traces
• Tower Middleware: Automatic instrumentation for MCP request handling

Quick Start

use turbomcp_telemetry::{TelemetryConfig, TelemetryGuard};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Initialize telemetry
    let config = TelemetryConfig::builder()
        .service_name("my-mcp-server")
        .service_version("1.0.0")
        .log_level("info,turbomcp=debug")
        .build();

    let _guard = config.init()?;

    // Your MCP server code here...
    Ok(())
}

Feature Flags

OpenTelemetry Integration

Enable the opentelemetry feature for full distributed tracing:

use turbomcp_telemetry::TelemetryConfig;

let config = TelemetryConfig::builder()
    .service_name("my-server")
    .otlp_endpoint("http://localhost:4317")
    .sampling_ratio(1.0)
    .build();

let _guard = config.init()?;

Prometheus Metrics

Enable the prometheus feature for standalone Prometheus metrics:

use turbomcp_telemetry::TelemetryConfig;

let config = TelemetryConfig::builder()
    .service_name("my-server")
    .prometheus_port(9090)
    .build();

let _guard = config.init()?;
// Metrics available at http://localhost:9090/metrics

Tower Middleware

Enable the tower feature for automatic request instrumentation:

use tower::ServiceBuilder;
use turbomcp_telemetry::tower::{TelemetryLayer, TelemetryLayerConfig};

let config = TelemetryLayerConfig::new()
    .service_name("my-mcp-server")
    .exclude_method("ping");

let service = ServiceBuilder::new()
    .layer(TelemetryLayer::new(config))
    .service(my_mcp_handler);

MCP Span Attributes

The telemetry system records MCP-specific attributes on spans:

Pre-defined Metrics

When using the prometheus feature:

License

MIT