mcplytics — Usage and Health for a Remote MCP Server

Open-source wrapper for a remote MCP server: which tools get called, which apps call them, how each call ended and how long it took. No SDK, no backend, and it never stores what people asked.

Open source · GitHub

Overview

In plain terms: you built an MCP server and put it online. AI apps now call it: Claude, Cursor, ChatGPT, scripts. mcplytics answers the plain questions about that traffic:

Which tools are used, and which never are?
Which apps call, and how often?
How many calls fail, and do they fail because a tool broke or because the request was wrong?
How fast is it? Median and slowest 5%.

It wraps the one function that answers your server's HTTP requests. You don't change your tools, add an SDK or run a backend. It never stores what people asked: no arguments, no answers, no IP addresses.

View on GitHub

What the report looks like

text

mcplytics: 21 calls, 2026-10-20 19:05 to 2026-10-20 19:05 UTC
failed 4.8% · p50 1 ms · p95 4 ms

Tools
name                 calls  failed       p50       p95
search                  12      0%      2 ms     11 ms
get_item                 6      0%      1 ms      2 ms
delete_everything        1    100%      0 ms      0 ms
list_categories          1      0%      2 ms      2 ms

Apps calling (client name)
name            calls  failed       p50       p95
claude-code         9     11%      1 ms     11 ms
cursor              7      0%      1 ms      4 ms
openai-mcp          4      0%      1 ms      4 ms
python-httpx        1      0%      2 ms      2 ms

A test run. delete_everything doesn't exist; someone tried anyway.

Install

bash

npm install mcplytics

On Cloudflare Workers, wrap your handler and bind an Analytics Engine dataset named MCPLYTICS:

import { mcplytics, analyticsEngine } from 'mcplytics';

const handler = async (request, env, ctx) => { /* your MCP endpoint */ };

export default { fetch: mcplytics(handler, { sink: analyticsEngine() }) };

On Deno, Bun or Node, write events to a local file instead:

import { mcplytics } from 'mcplytics';
import { jsonl } from 'mcplytics/node';

Bun.serve({ fetch: mcplytics(handler, { sink: jsonl('mcplytics.jsonl') }) });

Then read the report:

bash

npx mcplytics report mcplytics.jsonl --days 7

How a call can end

A tool that fails still answers with HTTP 200, so status codes alone can't tell you much. mcplytics reads the answer:

Outcome	Meaning
`ok`	a result
`tool_error`	the tool ran and failed (`isError: true`)
`protocol_error`	a JSON-RPC error: unknown tool, bad arguments
`http_error`	refused before any answer, e.g. 401 sign-in required
`no_response`	a stream that closed before answering
`crash`	your handler threw

The full list is in the README.

What it can't tell you

App names are what apps send. A client names itself in the request, and any name can be claimed. Good enough for counting traffic, never for deciding who gets in.
Remote servers only. A local server talks to its app over stdio, with no HTTP request to wrap.
No intent. It doesn't ask the model why it called a tool. That would mean adding a field to every tool.

Tech Stack

Language: JavaScript, zero dependencies, web APIs only
Runs on: Cloudflare Workers and Pages, Deno, Bun, Node
Works with: any Streamable HTTP MCP server: the official SDK, another framework, or by hand
Storage: Cloudflare Analytics Engine, a JSONL file, or your own callback

← All work