[Feature]: Add per-tool metrics for MCP tool invocations

[rite7sh]

Requirement

As a Jaeger operator, I want metrics for each MCP tool call,so that I can understand tool usage, latency, and errors.

Problem

The jaegermcp extension does not expose metrics for individual MCP tools.

Current logging works at the JSON-RPC level. It tracks methods like tools/call, but it does not show which tool ran.

Because of this, we cannot see:

• which tools are used most
• which tools are slow
• which tools fail often

This makes it hard to debug and improve MCP behavior.

Proposal

Add per-tool metrics for MCP tool calls in jaegermcp.

Start with two metrics:

• request count
• latency

Each metric should include:

• tool name
• result (ok or error)

Design:

Wrap each tool handler at registration time.

Example:

instrumentTool(toolName, handler)

The wrapper will:

• count each call
• measure latency
• record result (ok or error)

Send metrics through OpenTelemetry to the backend. Keep the logic in one place. Do not change individual handlers.

Follow-up work can include:

• request and response size
• more result types
• protocol-level metrics

(the diagram is simplified for ease of understanding )

Open questions

• Should we use Jaeger metric names like requests and latency under jaeger.mcp?

• Do we also want metrics at the method level (tools/call)?

• Should we split result into handler error and protocol error?

• Should we also add request and response size, more result types, and protocol-level metrics in a later step? ( I think yes this can give better visibility into tool behavior.)

[rite7sh]

Hello, @yurishkuro i just wanted a maintainer 's POV on this whenever you can in free time, is it worth addressing?

[yurishkuro]

there is a PR for that - #8160

[rite7sh]

oh okay, thanks for pointing that out!