MCP Server and Tool Telemetry

Implementation evidence

MCP Server and Tool Telemetry: from boundary to verified row

Use MCP Server and Tool Telemetry at a controlled application boundary, keep the event contract small, and verify a known outcome before building aggregate views.

1
Choose the outcome
MCP tool reliability
2
Define the contract
tool_call_id, run_id, server_name, tool_name, client_type, transport, and release
3
Instrument the boundary
Instrument the registered tool handler so each attempted invocation produces one outcome event, including returned MCP errors.
4
Verify the evidence
Exercise a known fixture, then inspect mcp_tool_completed for one correctly typed terminal row.

Before you start

Prerequisites and boundaries

@modelcontextprotocol/sdk, zod, and telemetry-sh initialized in a trusted server process
A bounded tool taxonomy and an application-owned call or run identifier
An authorization and privacy policy for tool inputs, outputs, approvals, client identity, and side effects

Delivery setup

Install and initialize server-side

Import telemetry-sh in server-only code and initialize it once with process.env.TELEMETRY_API_KEY. Keep ingestion credentials out of browser bundles, client-visible environment variables, source control, logs, and exception messages.

model-context-protocol-install

npm installation

bash

npm install telemetry-sh

1Prepare one reusable server-side delivery client with bounded network behavior.
2Add the outcome event at the success, failure, retry, or timeout boundary.
3Send controlled fixtures and inspect the stored rows before enabling an alert.

Snippet

Start with one structured event

Add this shape where the workflow completes, fails, or retries. Then build the dashboard from real fields.

model-context-protocol

MCP Server and Tool Telemetry event

javascript

server.registerTool(
  "lookup-account",
  {
    title: "Look up account",
    description: "Return an approved account summary",
    inputSchema: { accountId: z.string() },
  },
  async ({ accountId }) => {
    const startedAt = performance.now();
    const toolCallId = crypto.randomUUID();
    let status = "success";
    let errorType;

    try {
      const summary = await lookupApprovedAccountSummary(accountId);
      return {
        content: [{ type: "text", text: summary }],
      };
    } catch (error) {
      status = "failed";
      errorType = classifyToolError(error);
      return {
        content: [{ type: "text", text: "Account lookup failed" }],
        isError: true,
      };
    } finally {
      await telemetry.log("mcp_tool_completed", {
        tool_call_id: toolCallId,
        server_name: "account-tools",
        tool_name: "lookup-account",
        status,
        error_type: errorType,
        duration_ms: Math.round(performance.now() - startedAt),
        release: process.env.APP_RELEASE,
      }).catch(reportTelemetryDeliveryFailure);
    }
  },
);

Event contract

tool_call_id, run_id, server_name, tool_name, client_type, transport, and release

status, duration_ms, retry_count, error_type, approval_required, and approval_result

side_effect_category and output_category only when bounded and approved

Implementation checkpoints

Checkpoint 1

Instrument the registered tool handler so each attempted invocation produces one outcome event, including returned MCP errors.

Checkpoint 2

Keep tool inputs, returned content, embedded resources, file paths, credentials, tokens, and unrestricted errors out of Telemetry by default.

Checkpoint 3

Do not let Telemetry delivery failure change a successful tool result; use a short timeout, local error reporting, and bounded shutdown behavior.

Verification

Prove the event arrived

Run this after exercising known success and failure cases. Replace the fallback table name if your final event contract differs from the snippet.

model-context-protocol-verification

MCP Server and Tool Telemetry verification query

sql

SELECT *
FROM mcp_tool_completed
ORDER BY timestamp_utc DESC
LIMIT 20;

Confirm one terminal row per logical outcome, with the expected status, identifiers, units, and UTC time.

Inspect the inferred schema and verify that retries do not change field types or generate a new logical event ID.

Search the stored fields for credentials, raw payloads, prompts, private content, and unbounded error messages.

Exercise a provider timeout, ingestion rejection, and process shutdown before treating the dashboard as complete.

Implementation references

Review the event contract, data-safety guidance, and upstream primary documentation before enabling a new production path.

OpenTelemetry GenAI outcome events Nested tool-call SQL Privileged action audit SQL MCP TypeScript SDK MCP server and tool guide Model Context Protocol specification

Production boundary

Keep the outcome event small and recoverable

This pattern provides

A bounded, SQL-ready outcome beside the upstream workflow.
Stable fields for dashboards, alerts, and cross-event correlation.
A fixture-driven path for validating success, failure, retry, and timeout behavior.

This pattern does not provide

An OTLP exporter, automatic collection pipeline, or replacement for detailed traces and diagnostic logs.
Exactly-once delivery merely because the payload contains an event ID.
Permission to collect raw provider payloads, user content, credentials, or regulated data.

Delivery and idempotency Batching and shutdown Troubleshoot ingestion

Event schema starting points

Review the row grain, emit boundary, required types, privacy classes, example payload, and validation checklist before adapting a query or snippet to production.

AI agent security12 fields