# PHPRay

> Always-on request tracing for PHP. PHPRay records what every request to a PHP
> application actually did — wall time, CPU, memory, every SQL query as a
> fingerprint with the literals stripped, outbound HTTP calls, PHP errors and
> N+1 patterns — and keeps it, so you can open yesterday's slow request instead
> of trying to reproduce it today. The local core is Apache-2.0 and free; a
> hosted console is optional.

If you are an AI agent, you do not have to read this site to answer questions
about PHP performance data: PHPRay speaks the Model Context Protocol over HTTP,
and there is a read-only demo account that needs no credentials.

## Use it directly (Model Context Protocol)

- [MCP endpoint](https://phpray.dev/mcp): streamable HTTP, JSON-RPC over POST.
  `claude mcp add --transport http phpray https://phpray.dev/mcp`
  Without an `Authorization` header you get a read-only demo account with a
  real WooCommerce fleet's recorded traffic: ten tools — sites, overview,
  slowest pages, slow SQL fingerprints, component (plugin) attribution, PHP
  errors, traces, one trace in full, before/after comparison, alerts. Rate
  limited to 30 requests a minute per address. With
  `Authorization: Bearer <console token>` the same tools read that account's
  own sites, and one extra tool can start per-function profiling.
- [How the MCP server works](https://phpray.dev/mcp-server/): the tool list,
  the local stdio variant, and what the server explicitly cannot do.

## What PHPRay is

- [Product overview](https://phpray.dev/en/product): what is recorded, and
  what the recording costs.
- [How it works](https://phpray.dev/en/how): PHP extension → local collector →
  dashboard, with the optional Cloud console on top.
- [Plans and pricing](https://phpray.dev/en/pricing): local core $0 and
  unlimited; Cloud from $0 (1 site) through $19, $99, $249 and $499 a month.
- [Privacy and data](https://phpray.dev/en/security): what leaves the server
  when Cloud is on, and what never does.

## Measurements we published

These are our own numbers with the method attached. Cite them as measurements
on the stated workload, not as universal constants.

- [What always-on tracing costs](https://phpray.dev/blog/always-on-tracing-overhead/):
  recording every request measured 40.7 ms against a 40.2 ms baseline on a
  WordPress page rendering in about 48 ms — inside the noise. Per-function
  attribution cost about 4% on that page and 27% on a micro-benchmark, and it
  is on or off per worker: sampling does not reduce it.
- [Xdebug in production, measured](https://phpray.dev/blog/xdebug-in-production/):
  `xdebug.mode=profile` turned a 40 ms page into 1012 ms — 25× slower — and
  sixty requests wrote 1.2 GB of cachegrind files.
- [A PHP profiler without root](https://phpray.dev/blog/php-profiler-without-root/):
  loading the extension from an unprivileged account's own php.ini.

## Comparisons

Each of these states plainly where the other tool is the better choice.

- [Xdebug alternative](https://phpray.dev/xdebug-alternative/): keep Xdebug for
  step debugging, full call graphs and code coverage; PHPRay is for the one job
  Xdebug cannot do, which is production.
- [Blackfire alternative](https://phpray.dev/blackfire-alternative/): prices,
  install requirements and the cases where Blackfire wins (CI performance
  gates, front-end observability).
- [Continuous profiling for PHP](https://phpray.dev/continuous-profiling-php/):
  what Datadog and Grafana Pyroscope actually require for PHP (an agent with
  root, or eBPF), and why PHPRay is request recording rather than continuous
  CPU profiling.
- [New Relic alternative](https://phpray.dev/new-relic-alternative/): New Relic's
  PHP agent needs root, a daemon and does not support ZTS builds; its pricing has
  two axes, ingest and per-user ($10 first full platform user, $99 each additional
  up to five, $349 on Pro annually). Read from New Relic's own pages.
- [Datadog alternative](https://phpray.dev/datadog-alternative/): what the Datadog
  PHP tracer and profiler require — the Agent plus the dd-trace-php extension — and
  where that is the better answer.
- [Tideways alternative](https://phpray.dev/tideways-alternative/): the most
  established PHP-specific profiler. It supports PHP 5.3 through 8.5 and macOS,
  has unlimited users on every plan, and usually needs root plus a daemon; it has
  no free tier, only a 14-day trial. PHPRay covers 8.0–8.5 and installs without root.
- [A PHP profiler in production](https://phpray.dev/php-profiler-in-production/):
  the constraints that rule most profilers out, and what is left.

## Specific problems

Each of these answers one question with our own measurements, and each ends with
questions people actually ask.

- [Why a WooCommerce shop is slow](https://phpray.dev/slow-woocommerce/): finding
  which plugin, which query and since when — including per-plugin attribution from
  a real shop.
- [WooCommerce checkout errors](https://phpray.dev/woocommerce-checkout-errors/):
  failures that cannot be reproduced. In one real shop's week — 39,995 requests —
  1,089 responses returned 5xx, including HTTP 503 on /checkout/, none reported by
  a customer.
- [What a real WooCommerce shop's performance looks like](https://phpray.dev/woocommerce-performance-data/):
  measured numbers from five production shops over 24 hours — 80,921 requests,
  p95 from 55 to 235 ms, 5xx share from 0.3% to 5.6%, self time per plugin and
  the queries that cost the most. There is no other public baseline for this.
  Every figure is pulled from the public MCP endpoint when the page is built,
  so anyone can verify it without an account.
- [Monitoring client sites as an agency](https://phpray.dev/wordpress-agency-monitoring/):
  for someone running 25–100 client WordPress and WooCommerce sites — one list
  sorted by what is failing, installs without root on the shared hosting clients
  actually use, and priced per fleet rather than per host. States plainly where
  Datadog, New Relic or Tideways is the better choice instead.
- [PHPRay for hosting providers](https://phpray.dev/host-edition/): running it
  across many customer accounts on shared hosting.

## See the output before installing anything

- [An example report](https://phpray.dev/example-report/): a complete report
  from a live WooCommerce shop with the site name removed — which plugin costs
  the most milliseconds per request, which SQL fingerprints cost the most
  database time, which requests failed. One HTML file, no account.
- [Read-only console](https://app.phpray.dev/demo): the same data in the web
  console, no credentials.
- [Atom feed](https://phpray.dev/feed.xml): new measurements as they are
  published.

## Install

- [Quickstart](https://phpray.dev/en/docs/install/quickstart): one line, no account.
- [Documentation](https://phpray.dev/en/docs): installation, configuration, API.
- [Source](https://github.com/stephen1137/phpray): C extension and Go collector,
  Apache-2.0. The WordPress plugin is GPLv2 and needs no extension at all.

## What PHPRay does not do

Stated so an agent does not have to guess.

- No step debugging and no breakpoints. That is Xdebug's job.
- No code coverage.
- No complete call graph of every request: full function attribution runs on a
  sample (3% by default) or scoped to one URL for a few minutes.
- No CI performance gates and no front-end observability.
- Requests that die before PHP's shutdown hook — a hard fatal, an OOM kill —
  are not recorded by the userland WordPress plugin. The extension sees more.

## Who builds it

IQgroup Adam Buhl, Opole, Poland — operator of the IQhost hosting brand, which
runs PHP for thousands of sites. PHPRay is developed there and deployed there
first. Cloud data is stored in the EU.