Everything you get with TrackMCP
See 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.
Prove the server is healthy
Everything an owner needs to answer one question: is my server doing its job right now?
Every call, ranked and trended
See which tools get called, how often, and how that shifts week over week. The ones pulling weight and the ones just sitting there both surface immediately.
- Call volume and share per tool
- Trends across any time range
- Break any metric down by tool
Know which tool is slow before your users do
Every tool gets a latency and error-rate profile. p50 for the typical call, p95 for the tail that quietly ruins agent sessions.
- p50 and p95 latency per tool
- Error rate trended over time
- Sort tools by slowest or flakiest
Catch the errors hiding inside a 200 OK
MCP returns tool errors inside a successful response. Your HTTP logs and APM show green while agents retry and give up. TrackMCP reads the payload and counts the real failures.
- Detects isError responses your logs miss
- Separates transport health from tool health
- Flags the exact tool and failure reason
Your APM sees the green 200. TrackMCP reads the payload and counts the failure inside it.
Open any call and see everything
Arguments in, result out, timing, client, and the error if there was one. The full record of a single call, without grepping a log file.
- Full arguments and result payloads
- Per-call timing and client
- Jump from a failure straight to its cause
See who uses it, and how
The adoption side of the story: which AI clients call you, where they drop off, and which tools actually matter.
See exactly who is calling your server
Split every metric by the agent behind it: Claude, Cursor, ChatGPT, or a custom agent. Know where your adoption actually comes from.
- Usage share per client
- Growth per client over time
- Filter any view by client
Follow one agent session end to end
Replay a session in the exact order calls happened, and see the step where agents stall or drop off before they finish the job.
- Full call sequence per session
- Drop-off at each step
- Completed vs failed outcomes
Find the tools nobody calls
The tools carrying your product and the ones no agent has ever touched, side by side. Stop maintaining code and schema clutter that nothing uses.
- Most and least used tools
- Zero-call tools flagged
- Adoption trend per tool
The layer that reads the data for you
Dashboards show numbers. TrackMCP tells you what they mean, and what to do about it.
Detect
Finds the tools failing silently and the ones no agent calls, without you writing a single query.
Diagnose
Explains the root cause in plain English: the schema mismatch, the retry loop, the slow dependency.
Project
Estimates the calls each fix wins back, so you know what to ship first.
send_email is your biggest leak. It failed 94% of the time this week. Agents tried three times, then gave up. Your schema wants to as an array, but most agents send a plain string. Accept a string too and you win back about 2,100 calls a week.
Fits the stack you already run
One line to install. Then send the data wherever your team already looks.
TypeScript SDK
Wrap the official MCP server in one line.
Python SDK
Same one-line wrap for Python servers.
Slack alerts
A message the moment a tool starts failing.
Webhooks
Push events into anything you already run.
REST API
Query your own metrics programmatically.
OpenTelemetry export
Send spans to your SIEM or warehouse.