elementor-live
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ELIVE_WP_URL | Yes | Base URL of the WordPress site; needed for open_editor({postId}) and open without arguments. | |
| ELIVE_CDP_URL | No | Attach to a running Chrome (e.g. http://localhost:9222) instead of launching one. | |
| ELIVE_WP_USER | No | WordPress user for wp_rest. | |
| ELIVE_HEADLESS | No | 1/true runs Chrome headless (CI only; you will not see the editor). | false |
| ELIVE_PROFILE_DIR | No | Chrome profile directory (keeps the WordPress login). | ~/.elementor-live/profile |
| ELIVE_WP_APP_PASSWORD | No | Application Password for wp_rest (Users → Profile → Application Passwords). |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| prompts | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| canvas_statusA | Report the state of the live Elementor editor tab: whether a session is connected, the editor URL, postId, document title and type, Elementor version, whether V4 atomic elements are enabled, the current user selection (element IDs) and lastSeq of the activity buffer. Call this first in every session; if connected is false or the call fails with not_connected/no_editor_tab, call open_editor. The first call may start the dedicated Chrome window. |
| open_editorA | Navigate the dedicated Chrome tab to the Elementor editor of a page and wait until the editor ($e) is ready. Pass either postId (requires ELIVE_WP_URL; opens /wp-admin/post.php?post=&action=elementor) or a full editor url. Returns the same status object as canvas_status. If WordPress shows a login page, ask the user to log in once in that Chrome window and call open_editor again. After opening, call get_canvas_skeleton to learn element IDs. |
| get_canvas_skeletonA | Return the compressed element tree of the open document: {id, elType, widgetType, label, atomic?, children} per node, plus the current user selection, nodeCount, recent user events (selections, edits, Alt+Space prompts, saves) and lastSeq. This is the only source of element IDs besides create results — never invent IDs. Pass since= to receive only newer events; without it the last 20 events are returned. includeSettings=true adds non-default settings per node (more tokens); use inspect_element for a single element instead. maxDepth limits nesting for very large pages. |
| inspect_elementA | Inspect one element by ID (from get_canvas_skeleton or a create result): non-default settings, controlsHint (the most relevant setting keys for this widget type), rendered bounding box in the preview iframe, an excerpt of computed styles (display, padding, margins, gap, font size, line height, overflow), visible text, whether it overflows horizontally, its parent chain and direct children. Use it before update_settings to see the current values and their exact shapes; call get_widget_controls when you need a key that is not in controlsHint. Fails with element_not_found for unknown IDs. |
| list_widgetsA | List the widget types registered in this Elementor installation as [{type, title, categories, atomic}]. Use filter (case-insensitive substring of type or title) to search, e.g. "button" or "image". Use the type as widgetType in a create op with elType "widget". Prefer classic widgets (heading, text-editor, button, image, icon-list …) inside "container" elements; atomic types (prefixed "e-") are Elementor V4 elements and may reject classic settings. |
| get_widget_controlsA | Return the control definitions (setting keys) of a widget type or elType, e.g. "heading", "button" or "container": name, type, label, default, options (capped at 30), section, tab, responsive. Call this before writing settings you have not seen in inspect_element — never guess setting keys. Responsive controls accept suffixed keys (name_tablet, name_mobile). Value shapes follow the control type: dimensions {unit, top, right, bottom, left, isLinked}, sliders {unit, size}, colors as hex strings, selects as one of the option keys. |
| mutate_canvasA | Apply a batch of operations to the open document through Elementor's own command API, in the user's editor tab, as ONE undo step titled after label (the user can press Ctrl+Z). Ops: create {parentId?, at?, elType, widgetType?, settings?} (omit parentId to append at document root; elType "widget" needs widgetType), update_settings {id, settings} (merges keys), move {id, parentId, at?}, delete {id}, duplicate {id}, set_custom_css {id, css} (Elementor Pro custom_css; "selector" refers to the element). IDs must come from get_canvas_skeleton or from earlier create/duplicate results (returned per op as id) — never invent them. Within one batch, "$N" (e.g. "$0") as id/parentId refers to the element created by op N of the same batch, so a container and its children can be built in one undo step. Prefer elType "container" with flex settings (flex_direction, flex_gap, flex_justify_content, flex_align_items) over legacy section/column. Check keys with get_widget_controls first. Atomic V4 elements (elType/widgetType "e-*") can be created, but may reject classic settings (code unsupported_atomic). A failing op does not stop the others; inspect results[] and verify visually with capture_vision. |
| save_documentA | Save the open document via Elementor's save commands. mode "draft" saves a draft/revision without publishing, "update" saves an already published page, "publish" publishes it. Only save when the user asked for it or confirmed the result; mutations are visible in the editor before saving. |
| capture_visionA | Take a pixel-accurate PNG screenshot of one element (id) or of the whole preview viewport (no id), as rendered in the live editor; the ghost cursor moves to the element first so the user sees what you look at. device switches the editor's responsive mode (desktop, tablet, mobile) before capturing. Use it after every meaningful mutation to verify the visual result instead of assuming it. Returns the image plus {box, bytes}. |
| design_lintA | Check rendered geometry for common layout problems: spacing off the grid (grid, default 4 px), children overflowing their parent, overlapping siblings, headings that wrap, horizontal scroll, tap targets smaller than 44 px, empty containers. With id only that element and its descendants are reported. Findings are suggestions: show them to the user (agent_say with id) and only fix what the user wants — do not silently override deliberate design choices. |
| agent_sayA | Show a short speech bubble from the agent inside the user's editor, anchored at an element (id, the ghost cursor moves there) or at the top of the canvas. Use it to explain what you are about to change, to point at lint findings, or to ask the user something while they watch the canvas. Keep messages under ~200 characters. mood tints the bubble: neutral, happy, thinking, warning, sassy. |
| set_hudA | Update the small heads-up badge in the editor top bar: score (e.g. a 0–100 design quality estimate), delta (change since the last update, shown as +/-) and a short note. Purely informational for the user; it does not change the document. |
| get_user_activityA | Return what the user did in the editor since a sequence number: selections (ids), edits (change events with the history title and ids), Alt+Space prompts (text), saves and navigations, plus lastSeq. Pass since= to get only new events; re-read get_canvas_skeleton or inspect_element for elements the user changed before mutating them, so you do not overwrite their edits. |
| wait_for_user_promptA | Block until the user presses Alt+Space in the Elementor editor and submits a prompt, then return {text, selection, seq} — selection holds the element IDs the user had selected, which is what "this" refers to in the prompt. Returns {timeout: true} after timeoutSec (default 120, max 600). Use it for tag-team work: finish a task, call agent_say to report, then wait for the next prompt in a loop. If your MCP client enforces a short request timeout, pass a smaller timeoutSec and call again. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| elementor_workflow | How to work on the live Elementor editor with the elementor-live tools. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| status | Current state of the live Elementor editor session (same as canvas_status). |
TDQS
Scored across 14 tools
Most tools target clearly distinct resources or actions (skeleton vs. single-element inspect, mutate vs. save, widgets vs. controls). There is mild overlap in the user-event area: get_canvas_skeleton also returns recent events, get_user_activity is a dedicated event feed, and wait_for_user_prompt surfaces Alt+Space prompts, so an agent could pick the wrong one. canvas_status and open_editor also both return the same status object, though the descriptions sequence them clearly.
The dominant pattern is snake_case verb_noun (get_canvas_skeleton, list_widgets, inspect_element, save_document, capture_vision, open_editor, mutate_canvas). A few deviate to noun_verb or noun_state forms (canvas_status, design_lint, agent_say, set_hud), but names remain readable and largely predictable. Minor inconsistency, not confusing.
14 tools is well within the healthy range and each has a distinct role: session setup, discovery, mutation, verification, and user interaction. Nothing feels padded; even the UX tools (set_hud, agent_say) serve the live-editor collaboration model.
The surface covers the full edit lifecycle: connect, discover widgets/controls, read tree and element detail, batch create/update/move/delete/duplicate/custom CSS, save, screenshot-verify, and lint. Gaps exist: no media library/upload tool (images must already exist as attachment IDs), no document/page-level settings or global styles access, no history/undo tool (relies on the user pressing Ctrl+Z), and no template save/insert.