elementor-mcp
Provides tools for reading and editing Elementor-built page structures, managing widgets, global design tokens, templates, and history through the Elementor plugin on WordPress sites.
Enables managing WordPress pages, page metadata, templates, media, and content, with support for multiple WordPress installations and WordPress capability-based permissions.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@elementor-mcp show me the page structure for the homepage"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Elementor MCP
An MCP server for working on Elementor-built WordPress sites — reading page structure, editing widgets, managing global design tokens, and doing it without breaking anything.
Elementor keeps its entire page layout as a JSON blob in a protected postmeta
key (_elementor_data). The WordPress REST API will not read or write it, which
is why generic WordPress MCP servers can list your pages but cannot touch what
is on them. This project closes that gap.
What you get
50 tools covering page structure, element surgery, widget schemas, global design, templates, history and site-wide search.
Live widget schemas. Control definitions are read from the target site's own Elementor registry, so the schema you get describes exactly the widgets that site has installed — core, Pro, and any third-party addon pack — instead of a hard-coded list that goes stale.
Safe concurrent writes. Every write carries the content hash the page was read at and is rejected if the page changed underneath. No silent overwrites.
Undo. Every write snapshots the layout first.
elementor_restore_snapshotrolls back the last change.Atomic batches. Multi-step restructuring applies as one operation, or not at all — a page is never left half-edited.
Multi-site. One server can front many WordPress installs, each addressable by name, each independently markable read-only.
Related MCP server: elementor-mcp-agent
Two ways to connect
The bridge plugin can be used two ways, and they are independent — pick whichever fits how you work:
1. WordPress serves MCP directly (recommended). The plugin exposes a
JSON-RPC 2.0 endpoint at /wp-json/elementor-mcp/v1/mcp — the same
"Streamable HTTP" transport most remote MCP servers use, running statelessly
(no session, no persistent connection, which is exactly how a PHP request
already works). Point an MCP client — including a claude.ai custom
connector — at that URL and authenticate with a WordPress application
password. Nothing to install beyond the plugin itself, and it works from
claude.ai on the web, desktop or mobile.
MCP client ──HTTPS──▶ WordPress
└── Elementor MCP Bridge (this repo)
├── /mcp (JSON-RPC 2.0, 49 tools)
├── /documents ┐
├── /kit │ plain REST, same logic
├── /templates │ the MCP endpoint calls
└── ... ┘2. A local Node server (src/), talking to the bridge's REST API. Useful
for development, for MCP clients that only support local stdio servers, or
when you would rather the element-tree logic run outside the PHP process.
MCP client ──stdio──▶ elementor-mcp (Node) ──REST──▶ WordPress
└── Elementor MCP BridgeBoth paths end up calling the same Elementor APIs — Document::save() (the
same path the editor itself uses, so sanitisation, revisions and CSS
regeneration all happen correctly) and the live widget registry for schemas.
The element-tree mutation logic (insert, move, duplicate, reorder, wrap,
batch) is implemented twice, once in each language, and kept in parity
deliberately: 236 PHP assertions and 41 TypeScript ones exercise the same
cases side by side, so a change to one is expected to show up as a matching
change in the other's test file.
Install
1. The bridge plugin
Build an installable zip and upload it through Plugins → Add New → Upload Plugin in wp-admin:
./scripts/package-plugin.sh
# build/elementor-mcp-bridge-1.2.0.zipThe script refuses to build if any file fails php -l, so a broken plugin
cannot be packaged.
Or install it directly:
rsync -a plugin/elementor-mcp-bridge/ user@host:/var/www/site/wp-content/plugins/elementor-mcp-bridge/
wp plugin activate elementor-mcp-bridge # or activate in wp-adminRequires WordPress 5.9+, PHP 7.4+ and Elementor 3.x or 4.x.
Confirm it is live:
curl -u 'user:app password' https://example.com/wp-json/elementor-mcp/v1/statusOn Windows, curl is an alias for Invoke-WebRequest and does not accept
-u. Use the bundled diagnostic instead, which checks reachability, the REST
API, plugin activation and credentials in order so a failure tells you which
one broke:
.\scripts\Test-Bridge.ps1 -SiteUrl https://example.com -Username admin -AppPassword 'abcd efgh ijkl mnop'Or force real curl with curl.exe rather than the alias.
2. An application password
In wp-admin go to Users → Profile → Application Passwords, add one named
elementor-mcp, and copy the generated value. This is not the account's login
password; it can be revoked independently.
The bridge checks a real WordPress capability on every route, so the agent can only do what that user could do by hand. Give it an account with the least role that covers your use — Editor is usually enough, Administrator only if you need to change site-wide design settings.
3. The server
npm install
npm run buildConfigure
Single site:
WORDPRESS_URL=https://example.com
WORDPRESS_USERNAME=your-user
WORDPRESS_APP_PASSWORD="abcd efgh ijkl mnop qrst uvwx"Several sites — set ELEMENTOR_MCP_SITES to a JSON array:
[
{ "name": "live", "url": "https://example.com", "username": "u", "appPassword": "..." },
{ "name": "staging", "url": "https://staging.example.com", "username": "u", "appPassword": "...", "readOnly": true }
]Variable | Purpose |
| Single-site credentials |
| JSON array or name-keyed object of site profiles |
| Path to a JSON file holding the same structure |
| Which profile tools use when |
|
|
| Per-request timeout, default 30000 |
Connecting a client to the WordPress MCP endpoint (recommended)
The endpoint is https://example.com/wp-json/elementor-mcp/v1/mcp.
claude.ai (web or desktop): Settings → Connectors → Add custom connector, paste the URL, and continue — no token to generate or paste anywhere. It runs an OAuth flow on its own: you're sent to your site to log in (if needed) and approve the connection, then straight back to claude.ai, connected. See Authentication for what that mints behind the scenes.
Any other MCP client: if it accepts a static token instead, use
base64(username:app_password) as either HTTP Basic or a bearer token:
echo -n 'your-user:abcd efgh ijkl mnop qrst uvwx' | base64To check the endpoint directly, either way works for a plain curl test:
curl -u 'your-user:abcd efgh ijkl mnop qrst uvwx' \
-X POST https://example.com/wp-json/elementor-mcp/v1/mcp \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Connecting a client to the local Node server
Claude Code:
claude mcp add elementor -- node /path/to/elementor-mcp/dist/index.jsOr by editing an MCP client config directly:
{
"mcpServers": {
"elementor": {
"command": "node",
"args": ["/path/to/elementor-mcp/dist/index.js"],
"env": {
"WORDPRESS_URL": "https://example.com",
"WORDPRESS_USERNAME": "your-user",
"WORDPRESS_APP_PASSWORD": "abcd efgh ijkl mnop qrst uvwx"
}
}
}
}Authentication
Two ways in, and which one applies depends on the client, not on you:
Direct — HTTP Basic or bearer, with an application password. Every route — REST and MCP alike — accepts a WordPress application password (Users → Profile → Application Passwords) sent as HTTP Basic:
Authorization: Basic base64(username:app_password)That is native to WordPress; no code in this plugin handles it. The MCP endpoint additionally accepts the same value as a bearer token —
Authorization: Bearer base64(username:app_password)— for MCP clients whose connector UI only exposes a single "token" field.
Same credential, two header shapes; there is no separate "MCP token" to
generate. This is what the curl examples in this README use, and it is
enough for the local Node server, direct REST calls, and any MCP client that
accepts a static token.
OAuth — for clients that require it. claude.ai's custom connector, on
both web and desktop, does not accept a static token: it runs a full
authorization-code flow with PKCE (RFC 6749 + RFC 7636) against the
connector's own origin, the same as any other OAuth-only remote MCP server.
The plugin implements that flow — GET/POST /authorize, POST /token, plus
the RFC 8414 and RFC 9728 discovery documents — entirely to satisfy this
requirement; it does not add any capability beyond what the direct route
already offers, and the local Node server never needs it.
You do not generate anything for this path yourself. Add the custom
connector in claude.ai with the URL https://example.com/wp-json/elementor-mcp/v1/mcp
and it drives the flow on its own: a redirect to your site to log in (if not
already) and approve access, then straight back to claude.ai, connected.
What gets minted behind the scenes is, again, an ordinary application
password — named Claude MCP (OAuth, <timestamp>) — so it shows up
individually in Users → Profile → Application Passwords and can be
revoked from there like any other, with nothing extra to track.
The redirect target is checked against an allow-list
(emcp_oauth_allowed_redirect_uris, defaulting to claude.ai's confirmed
callback) before any code is issued, so this cannot be turned into an open
redirect by a crafted link.
CGI/FastCGI
If direct authentication fails with a 401 and the credentials are correct,
the server is likely running under CGI/FastCGI, which strips the
Authorization header before PHP sees it. Add this to .htaccess above the
WordPress block:
RewriteCond %{HTTP:Authorization} ^(.*)
RewriteRule .* - [e=HTTP_AUTHORIZATION:%1]One site, or many
The WordPress-hosted MCP endpoint has exactly one site: the one it runs on.
Its 49 tools take no site argument — there is nothing to disambiguate.
The local Node server can front several WordPress installs from one process
(see Configure above), which is why its tools all accept an
optional site argument. If you only ever work on one site, this
distinction will not come up; it matters once you are managing more than one
install from a single MCP client.
How to use it well
The tool descriptions steer toward this, but the short version:
elementor_site_statusonce per site — versions, widget count, what the authenticated user may do.elementor_get_outlineto get element ids. Do not reach forelementor_get_page_treeunless you genuinely need every setting; outlines are a fraction of the size.elementor_get_widget_schemabefore setting unfamiliar widget settings. Control names are not guessable, and a wrong key saves silently with no visible effect — this is the single most common way Elementor automation goes wrong.elementor_batch_editfor anything multi-step.elementor_render_pageto confirm the result.
Responsive settings use _tablet and _mobile suffixes (padding,
padding_tablet, padding_mobile); the schema marks which controls are
responsive and spells out the exact keys.
Safety model
Concern | How it is handled |
Concurrent edits | Writes carry the hash read at load; a changed page returns 409 and the server retries once against fresh data |
Mistakes | Every write snapshots the layout first; |
Half-finished edits | Batches apply in memory and validate before a single write; a failed operation writes nothing |
Corrupt trees | Duplicate ids, widgets with children, and widgets missing |
Destructive operations | Permanent deletion and site-wide replace need |
Site-wide replace | Defaults to a dry run that reports every page it would touch |
Permissions | Every route checks a real WordPress capability for the authenticated user |
Stale CSS | Elementor's cached CSS is flushed after every write, so changes actually render |
To allow permanent deletes and committed site-wide replaces, add to
wp-config.php:
define( 'EMCP_ALLOW_DESTRUCTIVE', true );Tools
Full reference in docs/TOOLS.md. By area:
Sites —
elementor_list_sites,elementor_site_status,elementor_native_mcp_callPages — list, get, outline, create, duplicate, delete, render, update metadata, full tree
Elements — get, find, add widget, add container, update, move, duplicate, delete, reorder, wrap, batch edit, replace tree
Widgets — list, schema, element types, dynamic tags
Design — globals, set global colour, update globals, custom CSS, global classes, flush CSS
Templates — list, get, save, apply, export, import
History — snapshots (list/create/restore), revisions (list/restore)
Content — search, find and replace, widget usage, media list, image upload
Elementor's own MCP module
Elementor 4.3 ships a built-in MCP module, gated behind the WordPress Abilities
API and focused on v4 atomic features — global classes, variables, components,
compositions. Where it is present, elementor_native_mcp_call proxies straight
through to it, so you get those abilities alongside everything here.
It is not a replacement for this server: it requires Elementor 4.3+ with
optional dependencies installed, and it does not cover the classic
section/column/widget model that the overwhelming majority of live Elementor
sites are built on. elementor_site_status reports whether it is available.
Development
npm run verify # every check below, except docs and packaging
npm run build # compile the Node server
npm test # 60 TypeScript unit tests
npm run typecheck # source and tests
npm run smoke # boot the Node server and exercise the MCP handshake
npm run test:guard # drive the Node server against a mock bridge
npm run docs # regenerate docs/TOOLS.md from the running Node server
php tests/plugin-load.php # load the plugin, register all 33 REST routes
php tests/tree-mutators.php # 236 assertions: PHP tree mutators against the same
# cases as tests/tree.test.ts
php tests/mcp-endpoint.php # 51 assertions: full JSON-RPC round trips against an
# in-memory WordPress — initialize, tools/list, and
# real element edits through the hash-guarded write path
php tests/oauth-flow.php # 46 assertions: the full authorization-code + PKCE
# round trip, driven directly against EMCP_OAuth
./scripts/package-plugin.sh # build the installable zipFour PHP harnesses, in order of what they prove:
tests/plugin-load.php— the plugin loads and every route resolves to a callable handler and permission callback. Catches activation fatals before they reach a zip upload.tests/tree-mutators.php— the PHP element-tree mutators (insert, move, duplicate, reorder, wrap, batch) behave identically to the TypeScript ones, case for case.tests/mcp-endpoint.php— the JSON-RPC layer itself: protocol negotiation, tool discovery, and real tool calls against an in-memory post store, including permission denial, the destructive-operation confirm gate, and the snapshot-then-restore undo path. It stubs WordPress rather than Elementor, so it exercises the bridge's own fallback write path, notDocument::save()— that half still needs a real site.tests/oauth-flow.php— discovery metadata; the not-logged-in-redirects- to-login case; consent approve and deny; PKCE accepted with the correct verifier and rejected with the wrong one; a used code rejected on replay; a mismatchedredirect_urirejected; an untrustedredirect_urirefused as a page rather than ever becoming a redirect; and the token this endpoint issues successfully authenticating through the existing bearer-token shim with no changes needed there. It callsEMCP_OAuth's private handlers directly via reflection, since the real request router ends inexit.
tests/capability-guard.mjs runs the built server over stdio against
tests/mock-bridge.mjs, a stand-in bridge whose element registry, widget
registry and stored documents a test dictates. That reaches the behaviour that
depends on the site rather than on our code: refusing an element type the site
cannot render, surfacing a save warning, standing down when a site reports no
registry at all, and re-applying a write that raced another editor. It asserts
against the mock's write log rather than the wording, so a regression that
keeps the message and loses the guard still fails.
Licence
MIT. See LICENSE.
Available Tools
50 toolselementor_add_containerAdd a container or sectionA
Insert a structural element — a flexbox container (the modern default), or a section/column for pages built the classic way. Containers are what you put widgets inside. You can seed it with child elements in one call.
Inputs: postId (required), site, elType, children, +3 more.
elementor_add_widgetAdd a widgetA
Insert a new widget into a page. Pass the widget type (heading, image, button, text-editor, and so on — see elementor_list_widgets) and its settings. With no targetId the widget is appended to the page root; a widget must otherwise go inside a container or column, so use position before/after to place it beside an existing widget.
Inputs: postId (required), widgetType (required), site, position, +2 more.
elementor_apply_templateInsert a template into a pageA
Insert a saved template's layout into a page, with fresh element ids. Place it at the page root, or relative to an existing element with targetId and position. The page is snapshotted first.
Inputs: postId (required), templateId (required), site, position, +1 more.
elementor_batch_editApply many element edits at onceADestructive
Run a list of element operations against a page as one atomic change: insert, update, move, duplicate, delete, reorder and wrap. Operations apply in order, each seeing the result of the last. If any one fails the whole batch is abandoned and nothing is written. Prefer this over a series of single-element calls when building or restructuring a section — it is one read and one write instead of many, and it cannot leave the page half-changed.
Inputs: postId (required), operations (required), site, label.
elementor_create_pageCreate a pageA
Create a new Elementor-enabled page, post or template. Starts empty unless you pass an element tree. Created as a draft by default so nothing goes live unreviewed. For a full-width blank canvas set pageTemplate to elementor_canvas.
Inputs: title (required), site, slug, type, +4 more.
elementor_create_snapshotSnapshot a pageA
Take a named snapshot of a page's current layout. Worth doing before a large restructuring so you have one clearly labelled point to come back to.
Inputs: postId (required), site, label.
elementor_custom_cssRead or write site custom CSSADestructive
Read the site-wide custom CSS held on the Elementor kit, or replace it. Passing css writes it; omitting css just reads. Note that Elementor only renders this on Pro.
Inputs: css, site.
elementor_delete_elementDelete an elementADestructive
Remove an element and everything inside it from a page. A snapshot is taken first, so elementor_restore_snapshot can undo it.
Inputs: postId (required), elementId (required), site.
elementor_delete_pageDelete a pageADestructive
Move a page to the trash, or delete it permanently with force. Permanent deletion additionally requires the site to have opted into destructive operations and confirm to be true.
Inputs: postId (required), site, force, confirm.
elementor_duplicate_elementDuplicate an elementA
Copy an element and its whole subtree, placed immediately after the original, with fresh ids throughout. The quickest way to repeat a card, column or row you have already styled.
Inputs: postId (required), elementId (required), site.
elementor_duplicate_pageDuplicate a pageA
Copy a page, its Elementor layout and its page settings into a new draft, regenerating every element id so the copy is independent. The usual way to iterate on a live page safely: duplicate it, edit the copy, then publish when it is right.
Inputs: postId (required), site, slug, title, +1 more.
elementor_export_templateExport a template as JSONARead-onlyIdempotent
Export a template in Elementor's own JSON format, ready to import into another site with elementor_import_template or through the Elementor UI.
Inputs: templateId (required), site.
elementor_find_elementsFind elements within a pageARead-onlyIdempotent
Search one page for elements by widget type, by text in their settings, or both, and get back their ids with a short label. Use this to locate what you want to edit without reading the whole tree.
Inputs: postId (required), site, text, limit, +1 more.
elementor_flush_cssRegenerate Elementor CSSAIdempotent
Clear Elementor's cached CSS so it regenerates on the next page load. The editing tools already do this after every write; call it manually when styles look stale after changes made outside this server, or after a theme or plugin update.
Inputs: site, postId.
elementor_get_elementGet one elementARead-onlyIdempotent
Read a single element and its subtree, including every setting. Use this after elementor_get_outline to inspect just the part of a page you are about to change.
Inputs: postId (required), elementId (required), site, settingsOnly.
elementor_get_global_classesGet v4 global classesARead-onlyIdempotent
Read Elementor v4 global classes — the reusable style classes used by the atomic/Editor One element model. Returns an empty list on sites that predate that feature.
Inputs: site.
elementor_get_globalsGet global design settingsARead-onlyIdempotent
Read the site's Elementor kit: global colour and typography palettes with the ids widgets refer to them by, layout defaults such as container width and breakpoints, and the site custom CSS. Read this before styling anything — using an existing global token keeps a site consistent, and changing one restyles every element bound to it.
Inputs: site, includeRaw.
elementor_get_outlineGet page structure outlineARead-onlyIdempotent
Read the structure of an Elementor page: every element id, its type, and a short label taken from its most identifying setting. This is the tool to reach for first when editing a page — it is a small fraction of the size of the full element tree and gives you the ids every other element tool needs. It also returns the content hash used for safe concurrent writes.
Inputs: postId (required), site, maxDepth.
elementor_get_pageGet a pageARead-onlyIdempotent
Read a page: metadata, a census of which widgets it uses, its content hash, and optionally the full element tree. Leave includeElements false unless you genuinely need every setting — element trees run to hundreds of kilobytes on real pages. Prefer elementor_get_outline, then elementor_get_element for the specific parts you care about.
Inputs: postId (required), site, includeElements.
elementor_get_page_treeGet the full element treeARead-onlyIdempotent
Read a page's complete element tree including every setting. This is the largest thing this server returns — only call it when you need to inspect settings across many elements at once. For normal editing use elementor_get_outline plus elementor_get_element.
Inputs: postId (required), site, summaryOnly.
elementor_get_templateGet a templateARead-onlyIdempotent
Read a template's element tree. By default ids are regenerated so the tree can be pasted into a page without colliding with anything already there.
Inputs: templateId (required), site, freshIds.
elementor_get_widget_schemaGet a widget's settings schemaARead-onlyIdempotent
Read the full control schema for a widget or structural element: every setting it accepts, grouped by panel tab and section, with types, defaults, allowed options and which settings are responsive. Read this before setting unfamiliar widget settings — guessing control names is the most common cause of an edit that saves without visible effect.
Inputs: name (required), kind, site, compact, +1 more.
elementor_import_templateImport a template from JSONA
Import an Elementor template JSON payload as a new template on this site. Pass the decoded JSON object from an Elementor export or from elementor_export_template — which makes copying a layout between sites a two-call operation.
Inputs: payload (required), site, title.
elementor_list_dynamic_tagsList dynamic tagsARead-onlyIdempotent
List the dynamic tags registered on the site — post title, ACF field, site logo and so on — which let a widget setting pull live data instead of holding a fixed value. Mostly an Elementor Pro feature.
Inputs: site.
elementor_list_element_typesList structural element typesBRead-onlyIdempotent
List the structural element types available (container, section, column) and the widget panel categories the site defines.
Inputs: site.
elementor_list_mediaList media library itemsBRead-onlyIdempotent
Search the WordPress media library and get back attachment ids and URLs. Elementor image controls need both, in the shape {"id":123,"url":"https://..."}.
Inputs: page, site, search, perPage, +1 more.
elementor_list_pagesList pages and postsARead-onlyIdempotent
List WordPress pages, posts and templates, newest edited first, with whether each is built with Elementor. Use elementorOnly to skip content the page builder does not own.
Inputs: page, site, type, search, +4 more.
elementor_list_revisionsList WordPress revisionsARead-onlyIdempotent
List a page's WordPress revisions, flagging which of them actually carry Elementor layout data. Revisions reach further back than snapshots but are less reliable for Elementor content.
Inputs: postId (required), site.
elementor_list_sitesList configured sitesARead-onlyIdempotent
List the WordPress sites this server can reach, with which one is the default. Call this first when you are unsure which site name to pass to other tools.
Inputs: no named parameters.
elementor_list_snapshotsList page snapshotsARead-onlyIdempotent
List the layout snapshots stored for a page, newest first, with when each was taken and why. One is taken automatically before every write this server makes.
Inputs: postId (required), site.
elementor_list_templatesList saved templatesBRead-onlyIdempotent
List the Elementor templates saved on this site — sections, containers, pages, headers, footers and popups. Reusing a proven template is cheaper and more consistent than rebuilding the same layout.
Inputs: page, site, type, search, +1 more.
elementor_list_widgetsList available widgetsARead-onlyIdempotent
List every widget type registered on the site, with its title, panel categories and whether it comes from Elementor Pro. Search by name or keyword to find the right one. Always check here before adding a widget — a site only has the widgets its installed plugins provide.
Inputs: site, search, category, includeHidden.
elementor_move_elementMove an elementA
Relocate an element within the page. Give a referenceId plus a position to say where it lands; omit referenceId to move it to the page root. Moving an element into its own descendant is refused.
Inputs: postId (required), elementId (required), site, position, +1 more.
elementor_native_mcp_callCall an Elementor native MCP abilityA
Invoke one of Elementor's own built-in MCP abilities through the bridge. Elementor 4.3+ ships abilities covering v4 atomic features (global classes, variables, components, compositions) that have no v3 equivalent. Only works where that module is active — check elementor_site_status first, under nativeMcp.
Inputs: tool (required), site, input.
elementor_render_pageRender page HTMLARead-onlyIdempotent
Render a page — or one element of it — to HTML as the front end would. Use this to check what an edit actually produced, especially for widgets whose output depends on dynamic data.
Inputs: postId (required), site, elementId, maxLength.
elementor_reorder_childrenReorder child elementsAIdempotent
Set the order of a container's direct children by listing their ids. Ids you leave out keep their relative order at the end, so you can move one item to the front without listing everything.
Inputs: order (required), postId (required), elementId (required), site.
elementor_replace_page_treeReplace a page's entire layoutADestructive
Overwrite a page's whole element tree. This discards the existing layout completely, so reach for it when generating a page from scratch rather than to make a change. A snapshot is taken first. For edits to an existing page use elementor_batch_edit instead.
Inputs: postId (required), elements (required), site, label, +1 more.
elementor_replace_textFind and replace across Elementor contentADestructive
Replace a string everywhere it appears in Elementor page data — a phone number, an old brand name, a changed URL. Runs as a dry run by default and reports exactly which pages and how many occurrences it would touch; set dryRun false and confirm true to commit. Every page it edits is snapshotted first. Always read the dry run before committing.
Inputs: search (required), site, dryRun, confirm, +4 more.
elementor_restore_revisionRestore a WordPress revisionADestructive
Restore a page's Elementor layout from a WordPress revision. Only works on revisions flagged as carrying Elementor data. A snapshot is taken first.
Inputs: postId (required), revisionId (required), site.
elementor_restore_snapshotUndo: restore a page snapshotADestructive
Roll a page back to a snapshot. With no snapshotId this restores the most recent one, which undoes the last change made through this server. The current state is snapshotted first, so a restore is itself reversible.
Inputs: postId (required), site, snapshotId.
elementor_save_templateSave a templateA
Save a layout to the template library, either from an element tree you pass or by copying from an existing page (optionally just one element of it). Save a section once, reuse it everywhere.
Inputs: title (required), site, type, elements, +2 more.
elementor_searchSearch Elementor content site-wideARead-onlyIdempotent
Search inside Elementor page data across the site — by text, by widget type, or by which elements define a given setting. Elementor stores its content as JSON in postmeta, so it is invisible to normal WordPress search; this is the way to find "which page has that old phone number on it".
Inputs: site, text, limit, postIds, +4 more.
elementor_set_global_colorSet a global colourAIdempotent
Change one global colour by id (primary, secondary, text, accent) or by its title, or add a new custom colour if it does not exist. Every element bound to that token updates at once, and Elementor's CSS is regenerated. This is the right way to restyle a site's palette.
Inputs: id (required), value (required), site, title.
elementor_site_statusCheck site statusARead-onlyIdempotent
Report what a site is running: WordPress and Elementor versions, whether Elementor Pro is present, how many widget types are registered, the active kit id, the capabilities of the authenticated user, and whether the bridge plugin and destructive operations are enabled. Run this before a first editing session on an unfamiliar site.
Inputs: site.
elementor_update_elementUpdate element settingsAIdempotent
Change an element's settings. In merge mode (the default) the keys you pass are applied over the existing settings and everything else is left alone; in replace mode the settings object is swapped wholesale, which discards any setting you do not include. Merge is almost always what you want.
Inputs: postId (required), settings (required), elementId (required), mode, +1 more.
elementor_update_globalsUpdate global design settingsADestructive
Merge settings into the Elementor kit — colour and typography palettes, container width, breakpoints, theme style defaults. This changes the whole site at once, so read elementor_get_globals first and change only the keys you mean to. A snapshot of the kit is taken before the write.
Inputs: settings (required), site.
elementor_update_page_metaUpdate page metadataAIdempotent
Change a page's title, slug, status, parent or featured image. This edits WordPress post fields, not the Elementor layout — use the element tools for that.
Inputs: postId (required), site, slug, title, +4 more.
elementor_upload_imageUpload an image from a URLA
Download a remote image into the WordPress media library and return the attachment id and URL, plus a ready-made value for an Elementor image control. Use this before setting any image setting — Elementor image controls want a library attachment, not a remote URL.
Inputs: url (required), site, title, altText, +1 more.
elementor_widget_usageReport site-wide widget usageBRead-onlyIdempotent
Count which Elementor widgets are actually used across the site and how much each page contains. Useful before removing an addon plugin, or to find the heaviest pages.
Inputs: site, postTypes.
elementor_wrap_elementWrap an element in a containerA
Put a new container around an existing element, in place. Useful for adding a background, padding or width constraint around something without rebuilding it.
Inputs: postId (required), elementId (required), site, wrapperType, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 50 tools
The read tools are explicitly layered — get_outline for structure, get_element for one subtree, get_page_tree for everything, get_page for metadata — and the descriptions state when to use which, which is unusually good for a set this size. A few clusters still require careful reading (batch_edit vs replace_page_tree vs duplicate_page, and snapshot vs revision restore), but no two tools are truly interchangeable.
Everything is snake_case with a uniform elementor_ prefix, and the vast majority is verb_noun (list_sites, create_page, update_element, restore_snapshot). Minor deviations are noun-only names like elementor_custom_css and elementor_widget_usage, plus the opaque elementor_native_mcp_call, but the pattern stays predictable.
Fifty tools is a heavy surface even for a broad site-building domain, and several groups are redundant in scope: four ways to read page content, five snapshot/revision tools, and both duplicate_page and duplicate_element. A capable agent can navigate it, but the sheer count raises selection cost and invites over-fetching of large trees.
Coverage is strong end to end: page and element CRUD, structural wrapping/reordering, atomic batch editing, templates, snapshots and revisions, globals, media, and site-wide search/replace. The notable gaps are lifecycle gaps on templates (no update or delete, only save/export/import) and no media deletion or update, which agents will hit when managing the template library.
Maintenance
Related MCP Connectors
Edit your live WordPress site from any MCP client: native Gutenberg blocks, rollback, prepaid.
Hosted WordPress MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Create, edit, preview, publish, and manage web pages from MCP-capable AI clients.
WordPress MCP server: publish posts, AI images, SEO and full site management, self-hosted
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceMCP server for AI-assisted WordPress editing across 12 page builders. 172 tools for content management, page builder editing, WooCommerce, SEO analysis, accessibility scanning, and site intelligence. Edits native builder formats (Elementor, Bricks, Divi, Gutenberg, Beaver Builder, and 7 more) with duplicate-before-edit safety, optimistic locking, and surgical element-level operations11MIT
- AlicenseAqualityCmaintenanceAgency-grade MCP server for WordPress Elementor — multi-site management for 120+ WordPress sites with safe edits (backup + auto-rollback + post-write verification), template export/import, global widget detection, CSS flush, WP-CLI escape hatch, and headless Chrome screenshots. 34 tools across pages, widgets, templates, bulk find/replace, and fleet operations.34101 npm3MIT
- FlicenseCqualityAmaintenanceMCP server that connects AI clients to WordPress for content management, site settings, Elementor editing, and more via a secure API.149-
- AlicenseNot gradedqualityBmaintenanceAn MCP server that enables AI agents to manage WordPress sites, including Elementor page building, content CRUD, plugin management, and site configuration via the WordPress REST API.MIT