Introduce Telemetry in Toolbox Core as per MCP Semantic Convention #582
Description
Prerequisites
- Search the current open issues
What are you trying to do that currently feels hard or impossible?
Add observability to the toolbox-core Python SDK via OpenTelemetry, including tracing and metrics, and context propogation by client-server link following MCP Semantic Convention
Suggested Solution(s)
Spans: Introduce spans for SDK operations.
- Ensure spans follow proper naming and attributes.
Metrics: Add metrics to capture SDK usage and performance.
- Follow OpenTelemetry semantic conventions (
mcp.client.session.duration,mcp.client.operation.duration).
Trace Hierarchy: Ensure client spans are linked as parents of corresponding server spans.
Alternatives Considered
No response
Additional Details
This introduces observability from scratch, enabling:
- Consistent telemetry across Toolbox stack
- Enable toolbox users debugging and performance monitoring
- Compatibility with OTLP-compatible tracing and metrics backends