Skip to main content

​
Create a telemetry setup

1

Open the Telemetry dashboard

Navigate to mcpfy.ai/dashboard/telemetry and start a new setup.
2

Tell us how you're running your server

Pick the option that matches your situation:
  • I’m building with mcpfy-sdk
  • I wrote my own server and can edit its code
  • I’m running someone else’s server on my computer
This is the same three-way split as the Overview - the dashboard shows you setup instructions tailored to whichever one you pick.
3

Get your API key

MCPfy generates an API key for this setup. Copy it - you’ll use it as MCPFY_API_KEY in the TypeScript or Python install steps.

​
Waiting for the first event

Until your server sends its first event, the setup shows a “Waiting for the first event” state with your API key (hidden by default - reveal or copy it from there) and the same setup instructions from step 2. Nothing to do here except finish wiring up the SDK - the first event usually shows up within a few seconds of your server’s next request.
Lost the setup instructions? Every telemetry setup has a Setup Instructions button that reopens the same picker at any time.

​
Once data is flowing

The setup page opens into three tabs: Analytics, Requests, and Events.

​
Analytics

A single scrollable page, ordered roughly by how often you’d check each section. Use the date-range picker and tool filter at the top to scope it - the tool filter only narrows Overview and Results & Payloads, the other sections aren’t tool-scoped.

​
Overview

Is anything broken right now? A health-score badge, total calls, and error rate, plus health-score and call-volume trend charts.
Telemetry Overview section showing server health score, total calls, error rate, and health score and call volume trend charts

​
Connections & Clients

Who’s actually connecting - handshake health, which optional capabilities connecting clients support, and a breakdown of client name/version by session.
Connections & Clients section showing init failure rate, protocol versions seen, client capability adoption, and client distribution by name and version

​
Results & Payloads

What’s coming back - result quality over time.

​
Tools & Catalog Quality

Is this server well-built for an LLM caller? Declared vs. called tools, catalog coverage, which tools have never been called, plus a per-tool table with call count, error rate, cancel rate, P95 latency, and risk flags (e.g. destructive tools) - the same tool catalog graded at two granularities.
Tools & Catalog Quality section showing declared vs called tools, catalog coverage, never-called tools, and the per-tool table with error rate, cancel rate, P95 latency, and risk flags

​
Server Signals

What the server pushes to clients unprompted, without being asked.

​
Usage Patterns

Call volume by day and hour (UTC), for the selected range.

​
Requests

A raw, per-request view of network traffic for the linked server.

​
Events

A raw terminal/event log stream for the linked server.

​
Managing a setup

From the setup’s detail page:
  • Copy or reveal the API key at any time
  • Archive a setup you’re pausing (and unarchive it later)
  • Revoke a setup permanently - type confirm to confirm. Any server still using that key stops sending telemetry immediately, and this can’t be undone.

​
What’s next

TypeScript

Install and configure mcpfy-pulse for Node.js MCP servers.

Python

Install and configure mcpfy-pulse for FastMCP or the official mcp SDK.