WP REST API Explorer — Manual

The WP REST API Explorer in detail: proxy setup, the check run and request limit, logins, all twelve tabs from Quick Check to Snapshots, and the limits.

Back to the overview: WP REST API Explorer · Open the tool: www.jpkc.com/tools/wp-api/

This manual documents the Explorer in full: how you connect the local proxy, how a check run works and what each of the twelve tabs shows. The tool's interface is in English, so tab, card and button names appear here exactly as you'll see them in the live tool.

Architecture and privacy first

  • Only through your local proxy. Every request to the site you check goes through localproxy on your own computer. This tool has no server-side proxy, and no Expert Mode switch like the other proxy tools either: without a connected proxy, the Explorer checks nothing. The tool page's Content Security Policy allows connections only to its own server and to 127.0.0.1/localhost; images from the checked site come in only through the proxy.
  • Your IP, your responsibility. Requests leave from your computer and your IP address. The tool says so plainly below the address field: Only check sites you own or are allowed to test.
  • Read-only. The transport sends nothing but GET and HEAD; the only _method override it allows is OPTIONS. Anything else it refuses before sending ("Not sent: this tool only reads (GET and HEAD).").
  • What stays in the browser. The browser stores the proxy address and token and shares them with the site's other proxy tools (Expert Mode). Credentials for the checked site are never stored — not in the browser, not in links, reports or exports.
  • Files are checked before they're saved. Every download is scanned for your own credentials and the proxy token. If either turns up, nothing is written ("Not saved: this file contains your …").
  • Foreign content stays contained. HTML from the checked site appears as text or in an isolated preview (sandboxed srcdoc, cleaned with DOMPurify). Images load only when you ask, through the proxy; the tool checks each file's header bytes, skips SVG and accepts at most 5 MiB per image.

Setting up the local proxy

At the very top sits the Local proxy card with a status line ("Local proxy: not connected", "Local proxy ready …") and the four-step guide. Once the proxy is ready, the guide collapses; Show setup guide brings it back.

Step 1: Download the proxy

Download the file for your system from the GitHub releases of the JPKCom proxy (free, open source, GPL-2.0-or-later): localproxy-windows-amd64.exe for Windows, localproxy-macos-apple-silicon or localproxy-macos-intel for macOS 13 and later, localproxy-linux-amd64 or -arm64 for Linux and WSL, plus builds for FreeBSD, OpenBSD and NetBSD. You need version 1.3.0 or newer; checks with a login and local sites need 1.4.0.

Step 2: Start it for this site

A switch shows the commands for Windows, macOS and Linux / WSL. The base command starts the proxy on port 8765, for the tool page's origin only:

./localproxy --origin "https://www.jpkc.com" --port 8765

On Windows it reads .\localproxy-windows-amd64.exe --origin "https://www.jpkc.com" --port 8765 (PowerShell), or the same without .\ in Command Prompt. --origin makes sure only this site may use the proxy; keep the window open while you work. The guide also covers the usual stumbling blocks: SmartScreen (More info → Run anyway, since the file is not signed), a start-localproxy.cmd for double-click starts (without it, a double-click starts the proxy on a random port and for every site), and Gatekeeper on macOS (chmod +x, remove the quarantine attribute, and if needed Privacy & Security → Open Anyway).

Next to it are the optional flags:

  • Local sites (1.4.0 and later): --allow-local "your-site.ddev.site" allows exactly the local site you're checking. The tool takes the host from the address field. --allow-local "*.ddev.site" opens every local DDEV project — sensible only if you may check all of them.
  • wp-env: --allow-local "localhost:8888".
  • Intranet: --dns system plus a network range, for example --allow-local "wp.intranet.example,10.0.5.0/24".
  • Large sites: --timeout 60.
  • DDEV over HTTPS: run mkcert -install once on the computer the proxy runs on.
  • Fixed token: --token-file "$HOME/.config/localproxy/token" (PowerShell $env:LOCALAPPDATA\localproxy\token, cmd %LOCALAPPDATA%\localproxy\token) keeps the token across restarts. Without it, the proxy creates a new one on every start.

According to the proxy documentation, allowed names reach loopback addresses only; a LAN target additionally needs an IP or CIDR entry, and without a port only 80 and 443 apply. Redirects from a public site into a local target are always blocked.

Step 3: Allow access to your local network

On first contact with the proxy, your browser may ask whether the page may access devices on your local network. Choose Allow. The request that triggered the question always fails — even after "Allow". The tool checks again by itself after a moment, or you press Check again. If you blocked it by mistake, the guide names the way back through the site settings in Chrome (Local network access) and Brave (Local network).

Step 4: Connect

From the proxy window, copy the Proxy address (default http://127.0.0.1:8765) and the Token (48 characters, 0–9 and a–f). The more convenient route: paste the proxy's entire startup box into Or paste the proxy's startup text — only the Address and Token lines are read. Then click Connect.

Connection states

The status line tells you what's wrong and offers the matching buttons (Check again, Copy start command, Enter the new token, Download the proxy):

  • Ready — with version and capabilities, for example "Local proxy ready · v1.4.0 · logins · local sites". With 1.3.0 it says "public sites only".
  • Unreachable — nothing answers. Is the proxy running, and did you choose "Allow"? After a double-click start without the .cmd file, the port is somewhere else.
  • Started for a different site — the proxy runs with a different --origin.
  • Token mismatch — "The token does not match. The proxy creates a new one on every start." Usually a restart.
  • Too old — a version below 1.3.0.
  • Not verified — from 1.4.0 on, the proxy has to prove cryptographically that it knows the token before the tool sends the token at all. If that fails, neither token nor password goes out. A 1.3.0 proxy doesn't know this proof: it gets the token for the public check, but credentials stay locked.

The "Site to check" card

Address and Check site

Enter the site's start address in Site address (https://example.com) and click Check site. If the proxy isn't ready yet, a click tells you which step of the guide is missing. If the site redirects within itself — from http to https, or between www and the bare domain — the tool checks the target address and notes it ("Checked … (the site redirects there)."). It does not follow redirects to another host or port.

Run settings

Available as soon as the proxy is ready:

  • Parallel requests — 1 to 4, default 3.
  • Rate (requests/s) — 0.5 to 4, default 2. Regardless of this, never more than 60 requests per minute.
  • Request limit — default 500, from 10 to 100,000. It counts requests to the site (one proxy call can be several, with redirects for instance), and it covers the whole check: the run and every click after it.
  • Content — Sample is the only active choice. Full download is greyed out ("Not part of the check yet"); you do full pulls per list in the Content tab.
  • Include probes (up to 11 requests) — switches the Security check's test requests on or off. Below it, "This check also sends these requests" lists every probe with its address.
  • Remember checked sites — keeps the checked addresses (without query) in the browser; Clear history empties the list, and so does switching it off.

Credentials

Locked until a proxy of version 1.4.0 or later is connected and verified. Two kinds:

  • Application password — User name and Application password (the one you create in the WordPress profile). WordPress accepts it only over HTTPS, local sites excepted.
  • JWT (paste a token) — a token you created elsewhere with a JWT plugin. The tool never sends your password to a token endpoint, since that would take a POST.

The order is fixed: first check the site without a login, then click Apply. Apply sends a single request to users/me and reports whether the login works; the next Check site then also runs with your login. Forget discards it. Credentials go only to exactly the site you entered, and only with read requests. Change the address, open a snapshot or reload the page, and they're gone. If the browser offers to save the password: Never.

Copy link copies a link with the address, the active tab and any search term. Whoever opens it only gets the address field filled in — it never starts a check on its own.

The check run

Phases

After Check site, the run strip "Check: …" appears with a phase badge and progress bar. The phases:

  1. Discovery — find the home page and the API root (via the Link header, /wp-json/ or ?rest_route=), read the index, run the diagnosis.
  2. Inventory — one HEAD request with ?per_page=1 per list (collection); the count comes from the X-WP-Total header.
  3. Sample — a sample of every automatically allowed list (around four requests each).
  4. Dates — date values for the charts; for large types, counts per year.
  5. Probes — the Security check's test requests.
  6. Evaluating — the assessment, without requests.
  7. Logged-in extras, Logged-in inventory, Logged-in sample — only with a valid login.
  8. Final probes — the two probes that can trigger an IP block, saved for the very end.

At the end it says "Finished"; empty tabs then point you to the Quick Check. Lists provided by a plugin or the theme are not queried automatically, because that runs their code — for those there are clicks in Overview and Content.

Request limit, pauses and blocks

  • Counter: the button "0 / 500 site requests" counts along and opens the request log. Details shows what the requests were spent on; with a login, split into with and without.
  • With a login, the limit is divided: 40 % for the public pass, 60 % for the logged-in pass.
  • Pause / Resume / Stop control the run. If the site asks for a pause via Retry-After, the tool waits with a countdown; above 60 seconds, the phase ends. If the site refuses several times in a row, the tool pauses: "The site has started refusing requests. The check is paused." — often a firewall blocking your IP.
  • Incomplete runs get a banner with the reason (stopped, limit reached, proxy gone) and the buttons Resume, Run remaining checks and Stop. If the limit is used up, raise it in Run settings.
  • Single clicks after the run may go past the limit — every button that sends a request names its cost, such as "(1 request)". Full pulls in Content, on the other hand, stop at the limit.

Request log

The log slides up from the bottom and lists every request with #, Method, URL, Status, ms, Bytes, Site req., Login and Step, filterable and sortable. One entry is one proxy call; a call can reach the site several times, so more site requests than entries may be counted. Entries that were not sent are flagged.

Partial results

When a tab is missing part of its data, it carries the badge partial. If the site gave no answer at all during the whole run, it says not checked.

The tabs

Quick Check

The starting point for everyone, sending no requests of its own. The Summary card opens with a headline sentence that is honest about coverage: "1 point needs action" or "{n} points need action", "We found nothing that needs action in the parts we could check.", "… in the content we looked at (…)" when it rests on a sample — and "Nothing needs action right now." only when the run was complete and everything is loaded. A broken front end is mentioned up front as well. Below that: the findings by level, the list of what wasn't checked, and the number of requests.

The Diagnosis card appears only when something stands out; otherwise the line reads "Site responds normally". Then come Security with the findings per level and Site at a glance with REST API, inventory, sample, up to five detected plugins, theme and WordPress version. Every card links to its tab; the discovery line ("API found at …") leads to Overview.

Overview

  • Site — Site name, Description, WordPress address, Home address, Time zone, Front page, Site logo, Site icon, Ways to log in to the API and API address (REST API root), plus the estimated WordPress version.
  • Namespaces — the groups of API addresses (from core, like wp/v2, or from plugins and theme) with their Kind and number of routes. Without a login you only see that a namespace exists; which plugin is behind it is shown in Plugins & Theme.
  • Content inventory — one row per list with Collection, Group, Route, Items, API pages (requests at 100 items per page), Status and Visibility. Only published content is counted. The notes underneath explain the special cases: "1 of 2 readable" (WordPress counts everything but shows visitors only part of it), "Not public (readable via API)" (marked as non-public but readable — seen from outside, hence "likely"), "Does not apply" and "Not available". Plugin lists are counted on click with Count (1 request).
  • Publication timeline — publications over time, for one type or all types compared.
  • Discovery — how the API was found.

Because anonymous responses may be cached by a CDN, a number can be slightly out of date; and because WordPress counts before it checks permissions, some counted items may stay unreadable for visitors.

Content

On the left the lists (a select field on small screens) with a badge such as "1 of 2 readable" where visitors can read only part of a list; on the right the items. Per list:

  • Sample (≈ 4 requests), Load all items (a full pull with Pause and Stop; the cost is shown up front) and, above the list navigation, Load all readable for every readable, automatically allowed list. Pulls count toward the request limit and stop there with part of the items. Lists in a plugin namespace are only queried on your click, because plugin code can have side effects.
  • Loaded data or Ask the site — filter what's loaded, or send a search to the site ("Send search (1 request)", up to 100 results, as a list of its own).
  • Filter, More filters (date range From/To), Items per page, Columns.
  • Four views: Visual (cards, a gallery for media), Table, JSON tree, Raw. Load images via proxy fetches the images.
  • Tick boxes or Select all put items on the export selection.

The item view opens at the side with Visual, Markdown (converted in the browser, without a request to the site), Fields, JSON tree and Raw. Links under the site's API open there as one request, other links open directly in your browser; Open in Structure jumps to the route.

The collapsible Charts section shows, depending on the list, a timeline, a heatmap by weekday and hour, freshness, authors or a tree for hierarchical types, and for media the files and file size per kind. Every chart states its basis (exact, or "based on N of M items") and can be read as a table with Show as table and copied with Copy as CSV. Buttons such as Load dates for all items, Count by year or counting the months of a year (12 requests) fetch missing data at the stated cost.

Structure

All routes from the REST index, as Tree or Table, filtered by text, Namespace, Kind, Method and Status. The status comes from the responses during the run: public, needs login, blocked or not requested. A route opens in Details, JSON tree or Raw:

  • Endpoints with methods and arguments; Load schemas of … (1 request) loads the schema for the whole namespace via ?context=help, or Ask this route (?_method=OPTIONS, 1 request) asks the route itself.
  • A request form built from the arguments: select lists for enums, "not set / true / false" for booleans, numbers as a text field. Blocked are password, _method, rest_route, and search on the user routes ("Not offered").
  • Snippets for curl, fetch and PHP (wp_remote_get) to copy; credentials appear in them only as placeholders.
  • Run (1 request) sends exactly one GET. For unknown plugin routes, the tool warns first and offers Run anyway (1 request). The response appears as a JSON tree or raw.

Tick boxes put routes on the export selection.

Three modes:

  • Loaded data — full-text search across everything loaded, without a request. In field narrows it to field paths (acf.* for all ACF fields, *.email), plus Type, Whole words, Search HTML source, Include routes (Structure) and Group content by type. Hits appear highlighted with their path, 25 per page; Filter list to hits and Add hits to export selection take them further.
  • Site search — one request to the site's /wp/v2/search ("Search the site (1 request)"); each further page costs one request.
  • One content list — the search of a single list. The tool never searches user lists.

WordPress only searches title, content and excerpt here, not meta or plugin fields. Results from the site end up under Loaded data afterwards.

Plugins & Theme

Without a login, WordPress gives out no plugin data. This tab shows what can be recognised from outside — the real number can be higher, because a plugin only shows up if it leaves a trace. Signals are REST namespaces, content types, fields, CSS classes in the markup, files under …/plugins/, …/mu-plugins/ and …/themes/ on the site's own host, ?ver= strings and generator tags. They're matched against a table of 1,010 entries (the 1,000 most popular plugins on wordpress.org plus namespaces from the wpprobe project and hand-maintained entries).

  • Plugins — as Cards, Table or JSON tree, with confidence (confirmed when signals from two sources agree, otherwise likely), signals and sorting. "One of several plugins" bundles signals that fit more than one plugin.
  • Unknown namespaces, Unknown types and Unknown taxonomies — traces the table doesn't know. The search link opens wordpress.org in a new tab; the tool itself queries nothing, and the routes of such namespaces are only available by click in Structure.
  • Must-use plugins — folders under …/mu-plugins/ from which the home page loads files.
  • Theme — from the body classes; Read theme details (1 request) reads the theme's style.css.
  • WordPress version — as "at least x.y", with where it was found.
  • With a login — with a login, WordPress lists all installed plugins (administrators), the active theme (any account that may write posts) and all themes (administrators). The column "Visible from outside?" then tells you, per plugin, what a stranger can still see of it.

Diagnosis

Explains why the site answers the way it does: home page, REST discovery, index and content, with a Verdict, a next step and "Also noticed" below. The Requests card lists the diagnosis requests; rows marked "evidence" carry the verdict. Verdicts are rated OK, Note or Problem, for example:

  • "Home page and API answer normally."
  • "Front end broken, API fine." — the theme or a plugin breaks the home page while the API runs.
  • "Displaying post content fails: a shortcode, block or plugin filter breaks."
  • "The API answers, but PHP messages come before the data." — the cause of the editor message "The response is not a valid JSON response."
  • "The path /wp-json/ is blocked by the server or firewall; the WordPress API itself works."
  • "API blocked (on purpose)." or "The API hints were removed from the pages (harmless hardening)."
  • "WordPress is in update mode." and "The whole site is password-protected by the web server (typical for staging)."
  • "Cloudflare's bot check stops automated requests. This is not a WordPress problem."
  • "Your IP may have been blocked by the site's firewall (user enumeration protection)."
  • "The API answers with a byte order mark (BOM) before the data." — browsers don't mind, PHP's json_decode does.
  • "This local address is not on the proxy's allow list." — for local sites, --allow-local is missing.

Security

What anonymous visitors can read through the API. The tab itself only sends something when you send a single probe on demand. Its cards:

Result — the count line by level, with no overall score, and the number of requests. The levels:

Level Meaning
Action needed Only with evidence that non-public data is exposed: other people's personal data (form entries, job applications) or secrets (keys, tokens). Always confirmed.
Worth reviewing Departs from WordPress defaults, or a protection that was evidently intended doesn't work, or only partly — without evidence of data exposure.
Good to know Default behaviour or something deliberately public that you should be aware of.
OK The check ran and found nothing, or the protection works.
Not checked The check didn't run; the reason is stated. Never counted as OK.

Plus the confidence: confirmed (seen in a response, and the evidence shows it) or likely (from a reliable but indirect signal such as index data or field names; the text then says "may" or "appears").

Findings — the 28 checks and their results, filterable by text, Level, Confidence and Has a snippet, as Cards, Table, JSON tree or Raw. The checks cover access to the API (logged-in users only? reachable? can a block be bypassed via ?rest_route=?), users (public list, an author named "admin", short names via the author redirect, the user sitemap and embed data, whether the site confirms that a user number exists), data (content types marked non-public but readable, keys, contact and bank details in fields that don't appear on the pages, password-protected entries), media (public documents, camera and author data in images), plugins (endpoints that might accept changes or submissions without a login, recognisable plugins) and files (readable WordPress version, application passwords, readme.html, XML-RPC, PHP file paths in error responses). Two of them run only with a login (see below). Every card shows coverage, evidence, "Why it matters" and "What you can do"; matches of sensitive values appear masked, and Reveal uncovers one value without sending anything.

Where a finding has a matching fix, Open the matching snippet: … opens the snippet in the WP Code Generator in a new tab:

Link in the Explorer Snippet in the WP Code Generator fits
Remove the WordPress version tag Remove WordPress version a readable WordPress version
Remove version query strings from assets Hide version query strings a readable WordPress version
Switch off XML-RPC Disable XML-RPC XML-RPC answers
Stop author redirects (?author=) Prevent user enumeration short names via the author redirect, or hidden names that are visible elsewhere
Hide the user list from the REST API Hide users REST endpoint from visitors a public user list, confirmed user numbers, a block bypassed via ?rest_route=
Remove the user sitemap Remove the users sitemap short names in the user sitemap, or hidden names that are visible elsewhere
Hide abilities from the REST API Hide abilities from the REST API abilities visible to accounts with few rights (login check)

Test requests (evidence probes) — up to eleven individual, anonymous GET requests without following redirects. They change nothing, count toward the limit and appear in the request log:

Probe Request What it shows
P1 …/wp/v2/users/999999 whether the site confirms that a user number exists (the response body is not kept)
P2 ?rest_route=… whether a block under /wp-json/ also covers the second address form (up to five, only when needed)
P3 ?author=<ID> whether the author redirect reveals a short name (not followed)
P4 readme.html whether the default readme is public
P5 xmlrpc.php whether XML-RPC answers (without calling a method)
P6 wp-sitemap-users-1.xml whether the user sitemap lists author pages
P7 oembed/1.0/embed?url=… whether the embed data names the author with a link

P3 and P1 run at the very end, because on sites that block scans for user names they can get your IP blocked; if the site asks for a pause before then, they're dropped. Never sent: forms, XML-RPC method calls, passwords or sequences of user numbers. Probes that weren't sent appear with "not sent: …" and a reason; Send this probe (1 request) sends a single one on demand.

Sensitive values in the loaded data — the pattern scan across everything loaded: e-mail addresses, phone numbers, IBANs (with checksum; sample IBANs from documentation don't count) and keys such as Stripe or AWS keys and private keys. The scan runs in a Web Worker without network access. Values are masked here, and reports always mask; the Raw view and an export with masking switched off show them in clear.

With a login — which parts were not checked without a login.

What this check does not judge — the deliberate limits: no lookup of known vulnerabilities (CVE) and no "outdated" verdict, no login, password or brute-force tests, no POST to xmlrpc.php, no writing routes, no search for files like .env, backups or debug.log, no rating of HTTP security headers or TLS (for that the tool points to the SEO & GEO Analyzer and DNS, SSL, Redirect & URL), no automatic calls to unknown plugin routes, nothing about the contents of uploaded files, no multisite network as a whole — and cached CDN responses can be out of date.

Reports

Two reports from the same data, as Markdown. They appear when the run ends; if you stop it, you get a report on what had loaded by then.

  • Technical report — for developers: requests, evidence, routes and field paths, with front matter and eleven sections (Summary, Scope and coverage, Not checked, Findings, OK, Probes, Site profile, Content inventory, Structure, Method and limits, Data export). E-mail addresses, phone numbers, bank details and keys are masked; author short names stay visible.
  • Customer report ("Website data check: …") — for the site owner: plain language, only counts and the affected content type, no routes, no names, no evidence. Sections: Summary, What we found, What is in order, What this check could not test, About this check. The headline sentence is the same as in the Quick Check.

View them as Preview or Markdown, then Copy, Copy without metadata or Download .md. The card What these reports cover sums up status and gaps and offers Run remaining checks and sending the open probes under "Still open". The Report pack bundles a README and both reports into a ZIP, optionally with data.json (Include data export (data.json)); Prepared by (for example "Example Agency") appears only in the customer report and is remembered in the browser.

Export

  • Format: JSON (one file with everything selected, as the site delivered it, plus findings and request log), CSV (one table per kind of content plus findings and log; several tables come as a ZIP with README), Markdown for AI (the data as text for AI tools, not the report; text from the site stays marked as a quotation), ZIP (everything ticked, with a README and a SHA-256 checksum per file) and OpenAPI 3.1 (a description of the API for tools such as Postman or Hoppscotch, only when a REST index is available).
  • Options: include the site's original answers; column separator comma, semicolon or tab; Make umlauts work in Excel (adds a BOM); keep HTML tags; include link and embed data; for Markdown, the full text converted in the browser; choose namespaces.
  • What to export: Everything loaded, Current view (exactly the rows of a list with its filter and sorting) or Selection (the ticks from Content, Structure, Security and Search). Sections such as Site info, Structure, Content, Media, Users, Findings and Request log can be deselected; data from the logged-in pass ("Logged-in data") is off by default.
  • Before you download: a file list with sizes, what the file is based on, Load content (≈ N requests) for missing content, and the column picker for CSV.

Passwords and keys are masked as far as the tool recognises them. Personal data such as names and e-mail addresses stays in the file. The switch Show passwords and secret keys in full lifts the masking — "Only for your own use. Do not pass this file on." Your own credentials are never written to a file. CSV files are protected against formula injection.

Snapshots

A snapshot is a JSON file with the data of a check that stays on your computer.

  • Save a snapshot — optionally with Include content (the text of each item; without it the file stays small, and changed text is still noticed, only the words are missing) and Include the site's original answers (larger; passwords and keys stay hidden unless you switch that on). Then Save snapshot.
  • Open a snapshot — choose a file or drop it in. It is checked first; after that, every tab shows the file and nothing is sent, not even to the proxy. A banner offers Leave snapshot; the request log shows the requests of the saved check. An applied login is forgotten when you open a snapshot.
  • Compare — two states of the same site: a snapshot against Current data or against a Second snapshot. The table shows Added, Removed, Changed and Not comparable per area; for findings, Added means "new", Removed "resolved", Changed "level changed". An item only counts as added, removed or resolved if both sides really saw it — otherwise it is "not comparable", with the reason. Save the result as Comparison as JSON or Comparison as Markdown.
  • Check these items re-checks missing items one by one (one request per item), Check the site again and compare fetches fresh data as the comparison side — the cost is on the button. Afterwards, Changed since snapshot in Content filters to the changed items.

Reference

A cheat sheet "As of WordPress 7.1": REST basics (namespace, route, endpoint, collection and item, _links, the API root via Link header, /wp-json/ or ?rest_route=/), authentication (anonymous, application password, JWT, cookie + nonce — the last one the tool cannot use), global parameters (context, _fields, _embed, per_page/page with a maximum of 100, _envelope, the headers X-WP-Total and X-WP-TotalPages), status codes (including 403 as a possible firewall page and 502/504 from the proxy rather than the site) and how to read the results.

Checking with a login

With a proxy of version 1.4.0 or later and an applied login, a second pass follows the public one: Logged-in extras (at most six requests in total: users/me and, depending on the account's rights, plugins, themes, active theme, settings and abilities — each only if the route is in the index), Logged-in inventory and Logged-in sample. The tool keeps the account's capabilities in memory only; files contain the role at most. The status line then reports "This check ran with your login · role: …".

Two security checks exist only this way: whether accounts with few rights can see plugin abilities, and whether users with a low role (such as Subscriber) can read data that visitors cannot. That's why a test account with few rights gives the most telling result; an administrator mainly shows the complete plugin list. The proxy sends credentials only to the exact origin, only with GET/HEAD, and over http only to local loopback targets.

Limits

  • Sample, not a full pull: the run loads a sample; a missing finding doesn't prove the whole site is clean. Full pulls happen in Content.
  • Seen from outside: plugins, theme and version are conclusions drawn from traces. "likely" means exactly that.
  • Not a vulnerability scanner: no CVE lookup, no attacks, no header or TLS rating.
  • Local sites only with --allow-local: without the flag, the proxy blocks private addresses.
  • The site's rate limits apply: the tool doesn't override Retry-After pauses or firewall blocks — it respects them.

For a first look, see the overview; concrete workflows are in the examples, strategy and pitfalls in the tips & tricks.