trackmcp

See how your MCP is being used

Track and analyze who uses your MCP server, what they are trying to do, whether the work gets done, and what to fix next. All from one line of code.

Get started
$ npm i @trackmcp/sdk

Add one line. Start learning

See who is using it
See what they use
Know what to improve

Illustrative workspace

app.trackmcp.com/overview

Overview

Live
Active clients+1
4
New connections+18%
1,284
Completed+24%
968
Completion rate+3.1pt
75%
Completed workflows
968 · last 7 days
MonTueWedThuFriSatSun
Traffic by client
Claude46%
Cursor31%
ChatGPT18%
Custom agent5%
Top tools
ToolCallsAdoptionStatus
search_docs14,208Healthy
create_issue9,841Healthy
run_query6,502Slow
get_user4,120Healthy

Works with every MCP client

ClaudeCursorChatGPTCustom agents
Understand usage

See who is using your MCP server

See which AI clients connect, how many are new, and how many keep coming back

acme-mcp-server Live
OverviewClientsToolsSessionsOutcomesReliability
128
Active clients
1,284
Sessions
62%
Completed workflows
1
Needs attention
What to look at
send_email needs attention — 94% failing
Checkout workflow stops early — 38%
Returning usage up 8% this week
Most common workflow412 sessions
Claude·completed
search_docsrun_querycreate_issue Completed
Cursor·stopped
search_docsrun_query Stopped here
4 in 5 sessions reach a useful result.
Workflows

See what people are trying to do

Follow the path from the first request to a useful result. See which tools people use, in what order, and where sessions stop

Outcomes

Know whether the work gets done

See which workflows reach a result, which ones need another look, and whether clients come back

Sessions to result75% completed
Sessions started0
Reached a tool0-19%
Completed a workflow0-7%
Returning clients
0%
come back within 7 days
Most completed workflow
search_docsrun_querycreate_issue
Most returning clients use more than one tool.
Reliability

Know where the job stops

Some calls look successful but never finish the task. TrackMCP shows where a request stalls, gets retried, or ends without the result the agent wanted

send_emailcall_4471

Visible response

200 OKRequest completed

Actual state

isError: trueSchema mismatch
Agent retries
Agent retries again
Agent gives up

Error summary

Tool
send_email
Error rate
94%
Avg retries
3.0
Primary issue
Schema mismatch

Suggested fix

Accept a string as well as an array.

One line of code

Add one line. That's it

Add one line to your server. TrackMCP takes care of the rest, and your data shows up in the dashboard right away

  • Works with the official TypeScript and Python SDKs
  • No manual event tagging
  • Data shows up in your dashboard in real time
$ npm i @trackmcp/sdk
server.ts
// wrap your existing MCP server
import { withTrackMCP } from "@trackmcp/sdk"
import { server } from "./mcp"
 
export default withTrackMCP(server, {
  apiKey: process.env.TRACKMCP_KEY,
  service: "acme-mcp-server",
})
One line in. Here's what happens next.
How it works

What happens after one line of code

Agents call tools on your MCP server. TrackMCP captures every call, turns it into sessions and outcomes, then writes you the answer in plain English

1AI clients make tool calls
Claude1,204 calls
Cursor812 calls
ChatGPT470 calls
Custom132 calls
tool calls
2Flow through your MCP server
Your MCP server+ 1 line
withTrackMCP(server)
search_docsrun_querycreate_issuesend_email
captures every call
3TrackMCP captures and synthesizes
TrackMCPanalytics layer
Groups calls into sessions, attributes clients, finds failures
delivers
4You get the outcomes
Who's using it4 clients
Tools12 active
Sessions3,120
Silent failures1 found
The answer, in plain English

send_email fails 94% of the time — agents send a string, your schema wants an array.

Fix the schema
The observability stack

Logs recorded it. APM watched it. TrackMCP explains it

Server logs

Records what happened

APM & tracing

Shows technical performance

TrackMCP

Shows who is using the server, what they are trying to do, and what to improve

Actionable insights

Get the answer, not another dashboard

TrackMCP turns your usage data into a short explanation of what is working and what to improve, with a useful next step

See how it works
Weekly insight

send_email is your biggest leak. It failed 94% of the time this week. Agents retried three times, then gave up. Your schema expects to as an array, but most agents send a plain string.

Fix: accept a string too. Estimated recovery: about 2,100 calls per week.

3 tools are dead. list_repos, get_customer, deploy_service received zero calls this week.

Analytics for MCP servers

Make your MCP measurable

Add one line of code and start understanding how your server is being used

Get startedRead the docs