mcp-audit-log

Intercepts and logs MCP tool invocations with structured JSON output, capturing tool name, arguments, return values, and execution metadata. Implements schema-based validation to ensure logged data conforms to predefined audit formats, enabling downstream parsing and compliance verification without custom parsing logic.

Captures and serializes tool invocation arguments into structured audit records, handling complex nested objects, arrays, and non-JSON-serializable types (Buffers, Dates, custom objects). Uses a configurable serialization strategy to represent these types in audit logs while preserving semantic meaning for later reconstruction or analysis.

Measures and logs execution duration, latency percentiles, and performance metrics for each tool call, capturing wall-clock time from invocation to completion. Aggregates metrics across multiple calls to enable performance profiling and bottleneck identification without requiring external APM tools.

Captures tool return values and error states, logging successful results alongside error objects, stack traces, and failure context. Distinguishes between tool-level errors (returned error objects) and execution errors (exceptions), enabling comprehensive failure analysis and debugging without manual error handling in tool implementations.

Emits audit log entries as structured events that can be consumed by external systems via event listeners or streams, enabling real-time log processing without blocking tool execution. Implements a non-blocking event emitter pattern that decouples logging from tool execution, allowing subscribers to handle logs asynchronously.

Captures MCP-specific context metadata alongside tool calls, including resource URIs, request IDs, user/session identifiers, and server state information. Enriches audit logs with MCP protocol context to enable correlation of tool calls across distributed systems and multi-step workflows.