trackmcp
Blog

Notes on MCP analytics

Playbooks and deep dives on measuring the usage of your MCP server: adoption, workflows, silent failures, and what to fix.

Guide

Why your MCP server needs analytics

Server logs tell you a request happened. They don't tell you who called what, why it failed, or whether the user got what they came for.

Feb 12, 2026·6 min read
Engineering

The errors hiding inside a 200 OK

MCP returns tool errors inside a successful response. Your HTTP logs and APM stay green while agents retry and give up.

Feb 6, 2026·7 min read
Guide

The metrics that matter for an MCP server

Not call volume for its own sake. The handful of numbers that tell you whether your server is adopted, reliable, and useful.

Jan 29, 2026·6 min read
Engineering

Instrumenting the MCP protocol layer

How to capture every call without touching a single tool: wrap the transport, not the handlers.

Jan 22, 2026·8 min read
Product

Find the tools no agent ever calls

Half the tools most teams ship are never called. Dead tools are schema clutter, maintenance cost, and confusion for the model.

Jan 15, 2026·5 min read
Guide

How to measure MCP tool adoption

Adoption is more than call count. It's how many clients use a tool, whether that's growing, and how it fits into workflows.

Jan 9, 2026·6 min read
Guide

Reading MCP sessions like funnels

A session is a path from first request to result. Treat it like a funnel and the drop-off points become obvious.

Dec 18, 2025·6 min read
Engineering

Why agents fail on your tool schemas

The single most common silent failure: a schema that expects one shape while agents reliably send another.

Dec 10, 2025·5 min read
Guide

Logs vs APM vs MCP analytics

Three layers, three jobs. Why logs and APM leave a gap that agent traffic makes impossible to ignore.

Dec 3, 2025·6 min read
Guide

Understanding your MCP client mix

Claude, Cursor, ChatGPT, custom agents. Knowing who drives your usage changes what you build next.

Nov 24, 2025·5 min read
Product

Workflow completion rate: the north-star for MCP

One number captures whether your server actually helps: the share of sessions that reach a useful result.

Nov 17, 2025·6 min read
Engineering

How to cut agent retries on your MCP server

Retries are a tax on every session. Most come from a small set of avoidable, measurable causes.

Nov 6, 2025·5 min read
Engineering

Latency that matters for agents (p50 and p95)

Averages hide the calls that ruin sessions. Track p50 for the typical call and p95 for the tail.

Oct 28, 2025·5 min read
Product

Retention for MCP servers

One-time usage is easy. The signal that your server is genuinely useful is clients that come back.

Oct 20, 2025·5 min read
Guide

Instrument your MCP server in one line

You don't need an observability project to start. Wrap your server, deploy, and read the data the same day.

Oct 10, 2025·4 min read
Guide

Writing tool descriptions agents actually use

A great tool with a vague description is invisible to the model. Descriptions are UX for agents.

Oct 1, 2025·6 min read
Guide

What is MCP analytics?

A plain-English definition: MCP analytics is product analytics for your MCP server — who connects, what they call, and whether it works.

Sep 24, 2025·5 min read
Guide

What is the Model Context Protocol (MCP)?

MCP is an open standard that lets AI models call your tools and data through a consistent interface. Here's the mental model.

Sep 16, 2025·6 min read
Engineering

How to monitor an MCP server

Uptime and status codes aren't enough for agent traffic. Here's what to monitor on an MCP server and how to set it up.

Sep 8, 2025·7 min read
Engineering

MCP observability: logs, metrics, and traces

The three pillars of observability, mapped to MCP servers — and the fourth thing agent traffic needs that they don't cover.

Aug 28, 2025·6 min read
Engineering

How to debug MCP tool errors

A practical workflow for finding and fixing the tool calls that fail — including the ones your monitoring calls a success.

Aug 19, 2025·6 min read
Product

How to increase adoption of your MCP tools

If agents ignore a tool, the problem is usually discovery, not capability. A practical playbook to get your tools used.

Aug 11, 2025·6 min read
Engineering

MCP server security: redacting sensitive data

How to get analytics on your MCP server without letting sensitive arguments or results leave your process.

Aug 1, 2025·5 min read
Guide

How to track which AI agents use your MCP server

Claude, Cursor, ChatGPT, or a custom agent? Here's how to see your client mix and why it changes what you build.

Jul 22, 2025·5 min read
Guide

MCP server metrics: a monitoring checklist

A copy-and-keep checklist of the metrics worth tracking on an MCP server, grouped by adoption, outcomes, and reliability.

Jul 14, 2025·5 min read
Product

How to measure the ROI of your MCP server

Tie your MCP server to outcomes leadership cares about: completed tasks, retained clients, and reduced failure cost.

Jul 3, 2025·6 min read
Guide

Answer engine optimization for MCP tools

AEO isn't just for content. Making your tools easy for an AI to choose and call correctly is optimization for the answer engine.

Jun 20, 2025·6 min read
Product

How to A/B test your MCP tool descriptions

Tool descriptions are UX for agents. Here's how to test wording changes and prove which version gets tools used correctly.

Jun 9, 2025·5 min read
Guide

MCP vs REST API: how analytics differs

If you've measured a REST API, MCP will feel familiar but different. Here's what changes when your callers are AI agents.

May 28, 2025·6 min read