@listo-ai/mcp-observability

Automatically intercepts and logs MCP tool calls with full context including tool name, arguments, execution time, and response payloads. Integrates at the MCP server protocol layer to capture invocations before they reach business logic, enabling observability without code instrumentation in tool handlers.

Captures inbound and outbound HTTP traffic with configurable payload sanitization rules that automatically redact sensitive fields (API keys, tokens, PII) before logging. Uses pattern-matching and field-name heuristics to identify and mask sensitive data without requiring manual annotation of every endpoint.

Provides a structured event emission API that allows developers to log domain-specific business events (e.g., 'user_signup', 'model_inference_completed') with typed metadata. Events are validated against optional schemas and enriched with automatic context (timestamps, user IDs, request IDs) before transmission to telemetry backends.

Captures user interactions in web applications (clicks, form submissions, navigation events) and emits them as structured telemetry events. Integrates with DOM event listeners and browser APIs to automatically track user behavior without requiring manual instrumentation of every interactive element.

Provides a pluggable backend interface that allows telemetry events to be routed to multiple destinations (e.g., Datadog, New Relic, custom HTTP endpoints, local file storage) without changing application code. Implements a provider registry pattern where backends are registered at initialization and events are fanned out to all active providers.

Automatically generates and propagates correlation IDs (trace IDs, request IDs) across MCP invocations, HTTP requests, and business events to enable end-to-end tracing. Uses async context (AsyncLocalStorage in Node.js) to maintain context across asynchronous boundaries without requiring explicit parameter passing.

Automatically collects timing metrics for MCP tool invocations, HTTP requests, and custom code blocks, then aggregates them into percentiles, averages, and histograms. Metrics are computed in-process and included in telemetry events, enabling performance analysis without external metrics infrastructure.

Automatically captures uncaught exceptions and errors, including full stack traces, error context, and breadcrumb trails of preceding events. Integrates with global error handlers and promise rejection handlers to ensure errors are logged even if not explicitly caught by application code.