EP Host Check
Roughly nine out of ten “PageMotor is broken” reports turn out to be the host quietly blocking something, not PageMotor. PHP cannot write the compiled CSS, outbound SMTP is blocked so email never sends, an upload limit is too low for a theme import. EP Host Check turns that guesswork into a clear report.
It is one admin panel that runs a battery of checks against the host your site is actually on, and gives you a plain-English green, amber or red result. Every line says what it means and how to fix it, written for a site owner, not a sysadmin. It diagnoses and never changes a setting on your server.
It also watches for known security problems. It checks whether PHP, the server’s Linux kernel and its database have their security updates, whether your HTTPS certificate is close to expiring, and whether the disk is filling up. Since 1.5.0 it warns you unprompted, in your admin and by email, when the PHP running your site has a known critical security hole.
Since 1.10.0 it also watches for things you did not add: a new address on your domain, a catch-all DNS record, or a new folder beside your site. These are the first signs that someone else has been in your hosting account.
Published by ElmsPark Studio.
The verdict
Section titled “The verdict”At the top of the report is a single overall verdict, taken from the worst result across every check:
- Green, “This host runs PageMotor fully”. Every check passed. Nothing to do.
- Amber, “This host has limits worth knowing”. PageMotor will run, but some items could cause confusion later. Worth a quick read.
- Red, “This host will break PageMotor, fix or move”. One or more checks will cause real failures (saving designs, sending email, taking payments). Fix them with your host, or move to a VPS.
What it checks
Section titled “What it checks”Can PHP save PageMotor’s files (the big one)
Section titled “Can PHP save PageMotor’s files (the big one)”This replicates what PageMotor does when it compiles your design: it writes a temporary file into the active theme folder, reads it back, and deletes it, then does the same for the uploads and user-content folders. If any of them cannot be written, your design changes will appear to do nothing and uploads will fail. This single restriction is the most common cause of a “PageMotor is broken” report, and it is a host setting, not PageMotor.
Capability checks
Section titled “Capability checks”- PHP version. Fails below 8.1, warns on 8.1, passes on 8.2 and newer.
- PHP security patches (new in 1.4.0). Whether the PHP actually serving your site has the fix for known critical PHP vulnerabilities, starting with PHP2Shell (CVE-2026-109618 and CVE-2026-109619). Red when it is exposed, with the upgrade commands. Ubuntu’s own PHP keeps its version number when patched (a patched Ubuntu 24.04 still says 8.3.6), so for those builds the check reads the running PHP’s build stamp instead. Because it asks the running PHP, it also catches an update that was installed but never restarted. Since 1.8.0 it also covers problems PHP has announced before a fix exists, starting with CVE-2026-115560. For those it reads the setting PHP says to turn off, from the PHP serving your site. Amber while it is still on, green (noted as mitigated) once it is off.
- Server restart for security updates (new in 1.6.0). Ubuntu installs Linux kernel fixes by itself but never restarts the server, so the fixes can sit unused for months. This compares the kernel your server is running with the latest security kernel, and turns amber when a restart is needed. It never turns red. It knows Ubuntu 24.04’s standard kernel; for any other kernel it says there is no data rather than passing.
- Database security updates (new in 1.7.0). Whether your MariaDB or MySQL server is missing a security update, from the exact build Ubuntu’s packages report. Amber at most, because a database is rarely reachable from outside. It knows Ubuntu 24.04’s own MariaDB and MySQL packages, and Percona Server 8.4 (what CloudPanel installs) from 1.7.2. A database from another source shows “no data” rather than a pass.
- PHP extensions. mysqli, curl, mbstring, json, openssl, and gd or imagick are required; zip is flagged as a warning (only needed for zip imports).
- Blocked PHP functions. Reads the
disable_functionslist. Fails if a file-write function orcurl_execis switched off; warns if onlyfsockopenis (that just affects the SMTP self-test). - open_basedir. If a path restriction is set, confirms it includes your site’s folders.
Outbound connections
Section titled “Outbound connections”- Outbound HTTPS. A real short-timeout request to a stable endpoint. If it is blocked, Stripe payments, licence checks, AI features and web fonts all stop working.
- HTTPS certificate (new in 1.7.0). Amber three weeks before your site’s certificate expires, red in the last week or once it has. Automatic renewal normally happens a month ahead, so a warning here usually means renewal is failing. It reads the certificate from your server itself, because behind a CDN the public address only shows the CDN’s certificate. If your server does not answer for the site’s name on its own HTTPS port, it reads the public address instead, and the row says so.
- Outbound mail (SMTP). A short-timeout connection test to a known mail server on ports 587 and 465. It never sends anything. Many shared hosts block this, which presents as “PageMotor won’t send email” when it is purely a host restriction.
Resource limits and caching
Section titled “Resource limits and caching”- Upload size (
upload_max_filesizeandpost_max_size), warned below ~64M. - Disk space (new in 1.7.0). Amber below 3 GB or 5% free, red below 1 GB or 1% free, when saves, uploads and backups start to fail.
- PHP memory (
memory_limit), warned below 128M, with a note about SCSS-compile headroom. - PHP time limit (
max_execution_time), warned below 30 seconds. - Object cache (Redis / Memcached): informational, never required.
- PHP opcode cache (opcache): on is faster; off is a gentle warning.
MCP / Claude connection (v1.2.0, extended through v1.9.0)
Section titled “MCP / Claude connection (v1.2.0, extended through v1.9.0)”Seven checks that answer “why won’t my site connect to Claude?” in one panel, each mapped to a section of the PageMotor MCP troubleshooting guide:
- Claude sign-in clock (OAuth timezone). PageMotor stamps the 60-second sign-in code’s expiry in UTC but reads it back in server-local time. A server ahead of UTC therefore treats every code as expired the instant it is issued, and browser sign-in from Claude always fails while everything else looks healthy. Red if your server is ahead of UTC now; amber if daylight saving will push it ahead later in the year. The fix is one line:
date.timezone = UTC. - MCP token reaches PHP. Some servers (typically Apache running PHP as CGI/FastCGI) strip the Authorization header before PHP sees it, so a perfectly valid token arrives as anonymous. When you run this report through the API or MCP with your token, that request is itself the definitive proof the header got through, and the check goes green. From the admin panel it reports what it can see and tells you how to get the definitive answer.
- HTTPS as PageMotor sees it. Behind Cloudflare Flexible SSL or a TLS-terminating proxy, your visitors use https but PageMotor thinks it is on plain http, so every sign-in URL it advertises is built wrong and the connection dies. Red when the report can see that mismatch from your proxy’s own headers.
- OAuth discovery reachable. PageMotor serves its OAuth discovery documents as dynamic routes, and two quite different faults can stop Claude reading them. A stock nginx/CloudPanel/Plesk
.well-knownblock answers them from the filesystem with a 404 (guide §6d). Or your host’s bot protection decides an automated client is not welcome and answers with a CAPTCHA challenge page — an HTTP 202 carrying HTML where JSON should be (guide §6g, new in 1.3.1). They need opposite fixes, so since 1.3.1 the check tells them apart rather than giving one diagnosis for both: only a 2xx carrying an HTML body counts as a bot wall, so a genuine 404 still reports as the filesystem block it is. Failure rows now quote the response content-type as well as the status, because the content-type is what shows why the row is red.- Bot wall. Nothing you can edit will fix it: the filter sits in front of your web server, and your own browser is trusted so you will never reproduce the fault yourself. The row says so, and gives you support-ticket wording naming the paths to exempt — including the advice to quote the HTTP 202, not “404”, because a support agent who hears 404 goes looking for a missing file.
- Filesystem block. Since 1.3.2 the fix text matches the server you can actually edit. On Apache or LiteSpeed the check reads your webroot
.htaccess: if a.well-knownshort-circuit rule is the cause it names the offending line and gives you the one-line narrowing toacme-challenge/only; if there is no such rule, the block sits in a layer only your host controls, so you get ticket wording instead of an edit that cannot work. On nginx you get the vhost block to fix. Before 1.3.2 every host got the nginx advice, which is useless on shared hosting where there is no vhost to open. - Both faults can stack on one host — 404 from the inside, CAPTCHA from the outside — and fixing only one leaves sign-in broken.
- Claude sign-in link reaches PageMotor (new in 1.9.0, guide §6i). Claude’s sign-in link carries its own return address (
https://claude.ai/...). Some hosting firewalls, such as Plesk’s Web Application Firewall, block any web address that contains a link to another site, so you get the host’s 403 page instead of PageMotor’s “Connect an App” screen. The row sends your site a harmless test link of the same shape. Red when a firewall stops it, with the one rule to switch off. - MCP endpoint answers POST directly. Connector clients POST to the slash-less MCP URL; a redirect turns that into an empty GET and the handshake dies even though the URL “works” in a browser. On PageMotor 0.9.x the redirect is expected platform behaviour, and the row tells you the fix is simply using the trailing-slash URL in your connector.
- No plugin intercepts MCP authentication (new in 1.3.0, guide §6f). Pre-OAuth bridge plugins (the PageMotor Architect AI Claude bridge, the EP MCP Bridge) answer the MCP route themselves behind their own API key. Left active on a modern core, they hijack authenticated requests only: sign-in completes, Connected Apps rows appear, then every request dies with “Invalid or missing API key.” — while anonymous probes, and the other six rows here, all look healthy. The check fingerprints every active plugin’s own code (never just its name — PageMotor bundles a harmless core plugin with the identical “PageMotor Architect AI: Claude” name) and names the offender. Amber, with the fix: deactivate it unless you deliberately connect through it.
These rows are version-aware: on PageMotor 0.9.x (which has no browser sign-in) the OAuth-only checks report as neutral “Info” rows rather than failures, and the bridge row’s advice becomes “retire it when you upgrade” since a bridge may be a 0.9 site’s intended connection path. Info rows never affect the overall verdict.
Requirements
Section titled “Requirements”- PageMotor 0.11.5 or later for EP Host Check 1.4.0 and later. Sites on older PageMotor keep 1.3.9.
- EP Suite base class (bundled)
Installation
Section titled “Installation”- Download
ep-host-check.zip: this link saves the file straight away. Once the plugin is installed, new versions appear in your PageMotor admin under Updates. - Upload via Plugins → Manage Plugins. Activate.
- Open Plugin Settings → EP Host Check, then click Host Report to open it. The checks run automatically.
Using the report
Section titled “Using the report”- Read the overall verdict at the top, then scan the rows. Each has a coloured dot, a plain meaning, and, for amber and red items, a prominent fix.
- Act on the red items first (those break things), then the amber ones (those are limits worth knowing).
- Use Re-run checks after you or your host changes a setting, to confirm the fix took.
Security alerts (new in 1.5.0)
Section titled “Security alerts (new in 1.5.0)”You should not have to open a report to find out your server has a critical security hole. So EP Host Check runs its PHP security check once a day on its own, and tells you when it finds one.
- A warning in your admin. It appears across the admin with a link to the fix, and clears itself once PHP is updated. If you work on your site through Claude, Claude sees the same warning.
- One email to your site’s administrators, sent through your site’s own email settings, with the fix and the upgrade commands. You get one email per problem, not one a day. If the problem goes away and later comes back, you are told again.
- One email per server, not per site (new in 1.7.0). If several of your sites share a server and the same list of administrator addresses, the first site to notice emails you and the others note that it went. Every site needs 1.7.0 or later. It works where the sites share the server’s temporary folder; on hosting that gives each site a private one, each site emails for itself as before.
- Switching the email off. Open Plugin Settings → EP Host Check → Security alerts and set Email security alerts to Off. The admin warning stays either way.
When it runs. With EP Cron active, the check runs daily even when nobody signs in. Without it, the check runs the first time an administrator signs in each day, or when Claude connects to the site.
After you update PHP, open the report and press Re-run checks. The warning clears straight away instead of waiting for the next daily check.
If the email cannot be sent, the admin warning says so, with the reason, so a broken mail setup is never silent. It tries again at most every six hours.
Watching for unexpected changes (new in 1.10.0)
Section titled “Watching for unexpected changes (new in 1.10.0)”When someone gets into a hosting account, they rarely touch the website. They add things beside it: a new address on your domain, or a folder for a website of their own. Your site keeps working, so nothing tells you.
Three rows at the end of the report, under Unexpected changes, look for exactly that.
Addresses on your domain
Section titled “Addresses on your domain”Every time an address such as shop.example.com is given a security certificate, it goes on a public record. This row reads that record for your domain. Since 1.11.0 the list comes from ElmsPark’s update server, so it is quicker and newer.
- The first time, it lists the addresses that are there today, and where each one points. Read the list. If you recognise them all, click I recognise these.
- After that, it tells you when a new address appears.
Addresses your hosting sets up by itself are not listed.
Catch-all (wildcard) DNS
Section titled “Catch-all (wildcard) DNS”A catch-all record makes every name under your domain answer, including names nobody set up. Some services add one when you connect them. If your domain has one, the row is amber and says where it sends those names.
If you set it up on purpose, click I recognise this. You are told again only if it starts pointing somewhere else.
Check nothing relies on a catch-all before you delete it. Any address served through it stops working.
Folders in and beside your site
Section titled “Folders in and beside your site”Hosting control panels give each website its own folder. So a website added to your account shows up as a new folder.
This row remembers the folders in your site’s top folder. Where your hosting account has its own home folder, as on most shared hosting, it also remembers the folders one and two levels above your site.
A new folder turns the row amber. The row says so when the folder is named like another website’s address, and when a website of that name points at your server.
What happens when something new appears
Section titled “What happens when something new appears”- The row turns amber and names what appeared.
- A warning appears in your admin. If you work on your site through Claude, Claude sees it too.
- Your administrators get one email, unless you have switched security alert emails off.
Nothing that was already there on the first look is emailed about. It is shown for you to read.
If you did not add it
Section titled “If you did not add it”- Take screenshots first. Do not delete anything yet.
- Change the passwords for your hosting, your control panel and your domain account.
- Turn on two-step sign-in for each of them.
- Ask your host’s support when it was added, and from which login.
When it runs
Section titled “When it runs”The catch-all and folder checks run once a day, the same way the security alerts do.
Reading the certificate record can take half a minute, so it never runs while you wait for a page. It runs when EP Cron is active, or in the background when you open the report.
Domain to watch
Section titled “Domain to watch”Leave Domain to watch empty and the plugin watches your site’s own address and everything under it.
If your site lives on a sub-address such as shop.example.com and you also own example.com, enter example.com to watch the whole domain. The setting is under Security alerts.
For developers and AI clients
Section titled “For developers and AI clients”EP Host Check exposes a host-report action over the PageMotor API and MCP. An MCP client (or the /api/ endpoint) can read a host’s status headlessly and reason about it, returning the same verdict and per-check data the panel shows (statuses pass, warn, fail, and the neutral info). Useful for checking a fleet of sites, or for an AI agent diagnosing a deployment.
The security rows have stable ids: php_security, kernel_restart, db_security, tls_certificate and disk_space. A security alert also appears in PageMotor’s own list-notifications, owned by EP_Host_Check with the code php_security.
From 1.10.0 the report ends with three more rows: watch_cert_names, watch_wildcard_dns and watch_folders. They are never fail. A row can carry two extra fields. ack names a list the owner can accept (names, wildcard or folders). refresh: true means the certificate record is due a read. Their warning has the code watch.
Two actions go with them:
watch-refreshreads the certificate record now and returns the three rows. It is slow, so allow up to 40 seconds.host-reportnever reads the record itself.watch-acknowledgetakescheck(names,wildcardorfolders) and records that the owner recognises what that row lists. Only call it when the owner has read the list and said it is theirs.
Two details worth knowing from 1.2.0: calling host-report through the API or MCP with a Bearer token is itself the definitive Authorization-header pass-through test — if it answers at all, the token reached PHP, and the report says so. And other EP plugins can gate on MCP readiness via EP_Host_Check_MCP::can_mcp_ready($checks), which is true unless any mcp_* check hard-failed.
What EP Host Check cannot do
Section titled “What EP Host Check cannot do”- Fix anything automatically. It reports what it sees, and the fixes are changes you or your host make. The only things it stores are its own daily check result, its one email setting, and the small alert marker described below.
- See another server. It runs on your host and reports what that host can do. The only thing about your site that leaves it is your domain name, for the lookups described under Privacy.
- Say who added something, or how. It can tell you a new address or folder has appeared. Only your host’s own records can say which login made it.
- See inside your hosting account. It cannot see your hosting logins, FTP accounts, email forwarders or DNS settings. It sees what is public about your domain, and the names of the folders around your site.
- Stop anyone. It warns you. Someone who holds your hosting login could also remove the plugin, which is one more reason to act on a warning promptly.
- See your site from the outside. Every probe runs on your own server, so a host that trusts its own machines but challenges outside traffic gets clean JSON on the loopback and a green “OAuth discovery reachable” row while Claude is still blocked. This is a known limitation, documented since 1.3.1 and not detectable from inside the box. If your site reports green here but still connects from nowhere, check it from another machine with
curl -sI https://yoursite.com/.well-known/oauth-authorization-serverand read the content-type, not the status.
Built to run on broken hosts
Section titled “Built to run on broken hosts”The whole point of the plugin is to run on the hosts that block things, so it never fatals. Every probe is wrapped and every function is guarded. If your host has disabled fsockopen, that is reported as a finding, not a crash. Every temporary file it writes to test is cleaned up afterwards, even when a write is blocked.
The one file it deliberately leaves is the alert marker that lets sites on one server send a single email. Each marker counts for 30 days, then a new one takes over. The old file is left for the server’s normal clean-up of its temporary folder. Its name is a code and it holds only a time, so it tells other sites on the server nothing about yours.
Privacy
Section titled “Privacy”A few requests leave your server. The only thing about your site that is ever sent is your domain name, which is already public.
- Two connection tests check only whether a connection can be opened: to a web endpoint over HTTPS, and to a mail server on ports 587/465.
- Since 1.11.0, the “Addresses on your domain” row gets its list from ElmsPark’s update server,
updates.elmspark.com. To keep that list ready, ElmsPark looks your domain up in the public certificate record through SSLMate’s Cert Spotter service, twice a day, for as long as the plugin is active on your site and for 35 days after its last update check. - SSLMate is sent your domain name and nothing else. ElmsPark keeps your domain name and the list of addresses, which is public record. The update server’s log keeps the time and IP address of each request for 14 days.
- If the update server cannot answer, the plugin asks the public certificate record at
crt.shdirectly, as 1.10.0 and 1.10.1 always did. Thencrt.shsees your domain name and your server’s address. - Since 1.10.0, the watch also asks a public DNS service (Cloudflare’s, or Google’s if that fails) whether a name answers. Nothing else about your site is sent. Folder names never leave your server.
- Since 1.4.0, it downloads the current list of known security problems, and its signature, from
updates.elmspark.com. These are plain file fetches that send nothing but the request itself. - Since 1.7.0, the certificate check reads your site’s own certificate. It asks your server directly first, and only falls back to your site’s public address if that fails.
Since 1.6.0 that list is signed. EP Host Check uses it only when the signature checks out against a key built into the plugin; otherwise it uses the copy bundled with the plugin.
Security alert emails go only to your own site’s administrators, through your site’s own email settings. No report is shared anywhere else unless you choose to share it. No telemetry, no tracking.
Troubleshooting
Section titled “Troubleshooting””The write check is red but I’m sure the folder exists”
Section titled “”The write check is red but I’m sure the folder exists””The folder existing is not the same as PHP being allowed to write to it. On many shared hosts the web user cannot write inside the site folder. The fix is a host permissions change, or moving to a VPS where PHP can write to its own site.
”Outbound HTTPS or SMTP is red, but my site loads fine”
Section titled “”Outbound HTTPS or SMTP is red, but my site loads fine””Inbound traffic (visitors reaching your site) and outbound traffic (your site reaching out to Stripe or a mail server) are different directions, controlled separately. A host can allow one and block the other. The report is about the outbound direction, which is what payments, licensing and external mail need.
”Your session has expired” the moment the panel opens (PageMotor 0.11)
Section titled “”Your session has expired” the moment the panel opens (PageMotor 0.11)”Update to 1.3.3, which fixes it. On PageMotor 0.11 with an earlier build, opening EP Host Check raised that alert immediately, and clicking OK reloaded straight back into it — while the admin dashboard and Content carried on working normally. Nothing was wrong with your session. Both the plugin and PM 0.11 were adding the CSRF token header to the same request, and because a browser appends repeated headers rather than replacing them, the token arrived twice in one header and core rejected it before the check ran. 1.3.3 sets the header only when core has not already set it, on 0.10 and 0.11 alike.
The same double-set affects the shared EP Suite admin header (the language selector and brand-colour picker) across the whole suite on 0.11. The shared code is fixed here first; each other EP plugin picks it up as it is rebuilt.
”The restart or database row says there is no data”
Section titled “”The restart or database row says there is no data””Those rows know Ubuntu 24.04’s own kernel and database packages. On other systems, or with a database from another source, they say so instead of guessing. That is not a fault on your server.
”The warning says the alert email could not be sent”
Section titled “”The warning says the alert email could not be sent””Your site tried to email your administrators and its email settings refused it. The warning gives the reason. Fix the site’s email (or ask your host), and the next daily check sends it.
”The addresses row says the certificate record could not be read”
Section titled “”The addresses row says the certificate record could not be read””The public record is a free service and is sometimes slow or unavailable. The row says when the last try failed and tries again later. Nothing is wrong with your site. The other two rows do not depend on it.
”The folders row lists folders I have had for years”
Section titled “”The folders row lists folders I have had for years””That is the first look. The row cannot know which folders are yours, so it shows you what is there and asks. Read the list, then click I recognise these. From then on it only speaks up about a new folder.
”The catch-all row says the record belongs to another domain”
Section titled “”The catch-all row says the record belongs to another domain””Your site’s address sits under a larger domain that answers for every name, which is common on hosting platforms. That record is the platform’s, not yours, and there is nothing for you to check.
”Outbound mail shows amber on a shared host”
Section titled “”Outbound mail shows amber on a shared host””That is expected on many shared hosts: they block outbound SMTP. Use the host’s own mail, or move external mail (such as Mailgun) to a VPS. See the Mailgun email setup guide.
Changelog
Section titled “Changelog”1.11.0
Section titled “1.11.0”- “Addresses on your domain” is quicker, more dependable and more up to date. The row used to say “could not be read” on a fair share of tries, and the record it read could be weeks behind. It now answers in about a second and sees certificates issued the same day.
- One more quiet first look. The first time the row reads the fuller record, the addresses it adds are listed for you to review and nobody is emailed. From then on a new address is flagged and emailed as before.
- What is sent where has changed. Your site now asks ElmsPark’s update server for your domain’s list. ElmsPark looks your domain up for you, so the public service no longer sees your server’s address. The Privacy section above has the detail.
1.10.1
Section titled “1.10.1”- The first look now lists everything, or says what it left out. On hosting with a great many folders, the “Folders in and beside your site” row named some of them and did not say there were more. It now lists far more, and when a list is still too long it says how many are not shown. The “Addresses on your domain” row does the same.
1.10.0
Section titled “1.10.0”- A watch for things you did not add. Three new rows, under “Unexpected changes”, look for signs that someone else has been in your hosting or your domain’s settings. You get a warning in your admin and one email when something new appears.
- New row: “Addresses on your domain”. Every address under your domain that has been given a security certificate is on a public record. The first time, the row lists what is there for you to read. After that it tells you when a new one appears, and where it points.
- New row: “Catch-all (wildcard) DNS”. A catch-all record makes every unused name under your domain answer. If your domain has one, the row says so and where it sends them.
- New row: “Folders in and beside your site”. It remembers the folders in your site’s top folder. On hosting where your account has its own home folder, it also remembers the folders above the site. A new folder turns the row amber, and it says so when the folder is named like another website.
- “I recognise these”. One click accepts what a row lists, and the row turns green. Nothing already there on the first look is ever emailed about.
- New setting: “Domain to watch”. Leave it empty to watch your site’s own address. Set it if your site lives on a sub-address and you also own the main domain.
- These checks send your domain name to a public certificate record and to a public DNS service. The Privacy section of the documentation names both. They change nothing on your server.
- Finds a firewall that blocks the Claude sign-in. Some hosting firewalls block any web address that contains a link to another site. Claude’s sign-in link carries its own return address, so the firewall answers with an error page and PageMotor never sees it.
- New row: “Claude sign-in link reaches PageMotor”. It sends your site a harmless test sign-in link, the same shape Claude uses. Green means it got through. Red means a firewall stopped it, with the exact setting to change (in Plesk: Web Application Firewall, switch off one rule).
- Amber means something else answered the sign-in page instead of PageMotor, with how to check it in a browser.
- The test never signs anyone in and never registers anything.
- The PHP restart command now restarts PHP. When PHP needs a security update, the fix tells you to restart PHP-FPM. The command it gave,
sudo systemctl restart 'php*-fpm', matches nothing on current Ubuntu, so it quietly restarted nothing. It now readssudo systemctl restart 'php*-fpm.service', which restarts every PHP version. - Clearer advice when the server’s kernel is behind. The restart row used to say the kernel fixes were “installed but not running yet” even when they had not been installed. It now says so only when the server reports a restart is waiting. Otherwise it tells you to install the updates first, then restart.
- No more false disk space warnings on shared hosting. On shared hosting your site sits on the host’s own very large disk, so even hundreds of gigabytes free can be a small percentage of it. EP Host Check now goes by how much space is actually free, so a site with plenty of room shows green. A disk that is genuinely filling up still turns amber or red.
- Warns about PHP security problems that have no patch yet. Some PHP problems are announced before a fix exists, with a setting you can turn off in the meantime. EP Host Check now reads that setting from the PHP actually serving your site. If it is still on, the PHP security row turns amber and says exactly what to turn off.
- First one covered: CVE-2026-115560. PHP advises turning off
session.upload_progress.enableduntil its next release. PageMotor does not use this setting, so turning it off changes nothing on your site. - Turned it off already? The row stays green and notes the problem as mitigated.
- An amber warning shows in your admin. Security emails are still kept for the critical, red problems.
- Percona Server is now checked. CloudPanel and some other control panels install Percona Server instead of MySQL. The database row now checks Percona Server 8.4 against its latest security release, where it used to say there was no data.
- The right upgrade command for Percona. When a Percona server is behind, the advice now names Percona’s own packages.
- Security emails on shared servers keep arriving. On a server where each site runs under its own account (CloudPanel works this way), the one-email-per-server feature could stop a security email from going out for a problem it had already reported more than 30 days before, until the server restarted. It now emails again as intended.
- New check: database security updates. The report now says whether your MariaDB or MySQL server is missing a security update. Ubuntu’s database packages show their exact build, so this works without guesswork.
- New check: HTTPS certificate. Warns three weeks before your site’s certificate expires, and turns red in the last week or once it has expired. Automatic renewal normally happens a month ahead, so a warning here usually means renewal is failing.
- New check: disk space. Warns when the server is running low, and turns red when it is nearly full and saves, uploads or backups could start failing.
- One security email per server, not one per site. If several of your sites share a server and the same administrators, the first site to notice a problem emails you and the others note that it went. Each site’s admin warning still appears.
- New check: server restart for security updates. Ubuntu installs Linux kernel fixes by itself but never restarts the server, so the fixes can sit unused for months. The report now compares the kernel your server is running with the latest security kernel, and says when a restart is needed.
- You are told when an alert email could not be sent. If the email to your administrators fails, the admin warning and the report now say so, with the reason, instead of retrying silently. It retries at most every six hours.
- The list of known security problems is now signed. EP Host Check only uses the published list when its signature checks out against a key built into the plugin. An altered or unsigned list is ignored, and the copy bundled with the plugin is used instead.
- You are told about a PHP security hole without opening the report. EP Host Check now runs its PHP security check once a day on its own.
- A warning appears in your admin when the PHP running your site has a known critical security hole, and clears itself once PHP is updated. If you work on your site through Claude, Claude sees the same warning.
- Your site’s administrators get one email about each problem it finds, with the fix. You will not get the same email every day.
- The email can be switched off in EP Host Check Settings, under Security alerts. The admin warning stays either way.
- With EP Cron active, the daily check runs even when nobody signs in. Without it, the check runs the first time an administrator signs in each day.
- Re-running the report after you update PHP clears the warning straight away.
- New check: PHP security patches. The report now says whether the PHP running your site has the fix for known critical PHP vulnerabilities, starting with PHP2Shell (CVE-2026-109618 and CVE-2026-109619, rated 9.2, fixed in PHP 8.2.35, 8.3.36, 8.4.27 and 8.5.12).
- It reads the PHP actually serving your site, not the command-line copy. So it also catches PHP that was updated on disk but never restarted.
- It understands Ubuntu’s own PHP. Ubuntu backports fixes without changing the version number, so a patched Ubuntu 24.04 server still says PHP 8.3.6. The check tells patched and unpatched builds apart from the build stamp of the running PHP.
- New advisories arrive without a plugin update. The list of known problems is published alongside the plugin’s updates and read each time the report runs. If it cannot be reached, the copy bundled with the plugin is used.
- A red result gives the exact upgrade commands for a VPS. On managed hosting, send the report to your host.
- Requires PageMotor 0.11.5.
- Settings language menu. The language menu in this plugin’s settings now lists only the languages it is actually translated into, plus English, so you can no longer pick a language that changes nothing.
- The Claude sign-in clock check no longer fails healthy servers. It used to decide from the server’s timezone alone: any host east of Greenwich failed the row, which dragged the whole report to “This host will break PageMotor”, even when sign-in demonstrably worked.
- The problem it warned about was real on PageMotor 0.10, where the sign-in code’s expiry was read back in server-local time and a server ahead of UTC treated every code as expired the moment it was issued. PageMotor 0.11 fixed that in core, so on 0.11 the warning was firing on a bug that is no longer there.
- The row now tests the actual round trip and reports what happened, rather than predicting from the clock. The detail line shows its working, so you can see the measurement instead of taking the verdict on trust.
- Servers still running PageMotor 0.10 are still failed, because for them the problem is genuine.
- The suggested fix now leads with updating PageMotor, which removes the dependency on the server clock entirely. Setting
date.timezone = UTCremains as a fallback, with a warning to check nothing else on the site relies on local time first: some scheduling features do.
- Corrects the PageMotor 0.11.3 preparation shipped in 1.3.6. That release grouped this plugin’s API and MCP actions into families, but in a shape PageMotor 0.11.3 does not accept. On 0.11.3 the plugin would have registered none of its actions, and nothing on screen would have said so.
- This version uses the shape 0.11.3 expects, and keeps the earlier shape for sites still on an older PageMotor. One build serves both, so there is no order you have to update things in.
- Safe to install now. Nothing you can see changes.
- Prepares the plugin for PageMotor 0.11.3, which requires every plugin to group its API and MCP actions into named families. On that release an action declaring no family is refused silently, so without this update the plugin’s actions would stop being offered with nothing on screen to say why.
- Families exist so that an AI working on your site sees the related actions together instead of stopping at the first one it tries.
- Safe to install on your current PageMotor. No behaviour changed.
Released 1 September 2026.
host-reportno longer appears as an individual top-level tool in connected MCP clients. It carried a flag that promotes an action alongside PageMotor’s own dispatch entry points, and core reserves that surface for itself.- Nothing became unreachable. The action id, method, access tier, arguments and handler are unchanged.
host-reportis still discoverable throughlist-actions, described bydescribe-action, and invoked throughcall-action. Only a caller that invoked the promoted name directly as an MCPtools/callis affected; it should dispatch throughcall-actioninstead.
The plugin could never deliver its own updates. Fixed. The Updates screen has never offered an EP Host Check update to anyone, 1.3.3 included. The update channel was serving it correctly the whole time; the plugin was the problem.
PageMotor builds its update check from each plugin’s Updates: header and skips any plugin whose header does not carry one. This plugin’s header block never had that line, so core excluded it from every update check ever made. It was invisible to the update system rather than up to date. The header is now present, matching the rest of the suite.
This fix cannot deliver itself. An install already on 1.3.3 or earlier has no Updates: header, so it still cannot see this release. Download 1.3.4 and upload it once through Plugins → Manage Plugins; from then on the Updates screen works normally.
Fixes the “Your session has expired” loop on PageMotor 0.11: the CSRF token header was being set twice — once by the plugin, once by 0.11’s new automatic injection — and arrived carrying the token twice over, so core rejected the request before the check ran. The plugin now sets the header only when core has not already done so, with no version sniffing to go stale. Verified end to end on a real 0.11 site: before, 31 alerts in nine seconds and a panel stuck on “Running checks against your host…”; after, no alerts and the full report renders. What the plugin checks and reports is unchanged from 1.3.2.
The “OAuth discovery reachable” fix text now matches the server you can actually edit. The filesystem-block failure used to prescribe the nginx vhost block to everyone, which is right on a VPS, CloudPanel or Plesk box and useless on shared hosting where there is no vhost to open. The check now branches on your serving environment: Apache or LiteSpeed with a .well-known short-circuit rule in the webroot .htaccess gets the offending line named and a one-line fix; Apache or LiteSpeed without one gets paste-ready support-ticket wording, because the block is in a layer the host controls; nginx gets the original vhost text. The .htaccess scan is bounded and deliberately never blames a correctly narrowed ACME rule.
“OAuth discovery reachable” stopped giving one diagnosis for two different faults. A host bot wall answering the discovery documents with a CAPTCHA challenge was correctly failing the check, but was reported as the nginx .well-known block and prescribed its fix — advice that cannot work when the filter sits in front of the web server. There is now a separate branch for the bot-wall case (guide §6g) with support-ticket wording, kept deliberately narrow so a genuine 404 still reports as the filesystem block it is. Failure rows also quote the response content-type, so a red row shows why it is red. Found in the wild on SiteGround Anti-Bot AI. Known limitation, unchanged since: these probes run from your own server, so a host that trusts its own machines can produce a green row while Claude is still blocked from outside.
New check, “No plugin intercepts MCP authentication” (guide §6f): finds active pre-OAuth MCP bridge plugins that answer the MCP route behind their own API key. Sign-in completes and Connected Apps rows appear, then every request dies with “Invalid or missing API key.”, while anonymous probes and the other MCP rows all look healthy. Detection fingerprints each plugin’s own code rather than its name, because PageMotor bundles a harmless core plugin with an identical display name. Warn-only: it can turn the verdict amber, never red.
New MCP / Claude connection check group: sign-in timezone, header pass-through, HTTPS detection, discovery reachability and redirect behaviour, the whole “why won’t my site connect to Claude” class in one panel. Read-only and bounded, with a neutral info status for anything that cannot be verified from inside, so a host that blocks self-connections never produces a false fail. Also fixes a compatibility problem with PHP 8.5 that could corrupt the report on exactly the misconfigured hosts this plugin exists to diagnose.
The admin check routes through a skew-safe wrapper, so a version mismatch between EP Suite plugins can never break this plugin’s admin screens. No functional change.
PageMotor 0.10 compatibility: the admin guard used a check 0.10 removed, which failed closed and locked admins out of the plugin’s admin actions. Both 0.9 and 0.10 are supported. Adds the standard build script.
Feedback and corrections
Section titled “Feedback and corrections”For a quick question about this plugin, EP Support inside your admin is the fastest option. The chat widget sits on every EP plugin settings page and knows which one you’re on.
For anything bigger, a bug report, a feature request, or a “how do I…” that needs a real reply, open a ticket at help.elmspark.com. A real person, helped by AI, writes the reply, usually within a few hours.