Empty server and transport fields in Prometheus metrics #1898
Description
Bug description
Starting with v0.2.8, several fields in the Prometheus metrics scraped from the /metrics endpoint or collected via OTel are empty, which were previously populated.
An example:
toolhive_mcp_requests_total{mcp_method="tools/call",method="POST",otel_scope_name="github.com/stacklok/toolhive/pkg/telemetry",otel_scope_schema_url="",otel_scope_version="",server="",status="success",status_code="200",transport=""} 4
The server and transport properties are missing.
Steps to reproduce
Run an MCP server with the Prometheus endpoint enabled:
thv run fetch --otel-enable-prometheus-metrics-path
Perform any MCP action like listing the tools or calling a tool from a client. Then, grab the metrics:
curl http://127.0.0.1:<MCP_PORT>/metrics
Expected behavior
The server and transport fields should be populated in metrics records.
This was an example from v0.2.7:
toolhive_mcp_requests_total{mcp_method="tools/list",method="POST",otel_scope_name="github.com/stacklok/toolhive/pkg/telemetry",otel_scope_schema_url="",otel_scope_version="",server="fetch",status="success",status_code="200",transport="streamable-http"} 2
Actual behavior
But with v0.2.8 and above server and transport values are empty:
toolhive_mcp_requests_total{mcp_method="tools/list",method="POST",otel_scope_name="github.com/stacklok/toolhive/pkg/telemetry",otel_scope_schema_url="",otel_scope_version="",server="",status="success",status_code="200",transport=""} 2
Environment (if relevant)
- ToolHive version: 0.2.8 and above (noticed with v0.3.0)