Paraplex Mcp

Whether a Paraplex tab is attached, and if not, what to tell the person: in Paraplex, Menu → Connect an agent, then this port and code. Call it first; say the code to the person exactly.

無需輸入參數。

The built-in templates: id, label and categories (the first is its section on the home page, e.g. app-promo, brand-promo, starter, website-hero-visuals, titles-text, infographics).

無需輸入參數。

A blank 1920×1080 project in the attached tab (the scene being left is saved as the editor saves it).

無需輸入參數。

Apply a built-in template by id, as the gallery does.

The open scene: canvas, duration, current time, and every layer with its kind, name, clip and parent.

無需輸入參數。

Move the playhead to a time in seconds.

A picture of the frame at a time (or the current one), through the export path — the way to look at what you made.

The frame at several times (default: four across the piece), one picture each.

The assistant's tools — every verb that edits the scene — with their input schemas. Run one with run_tool.

無需輸入參數。

Run one of the assistant's tools in the attached tab (see list_tools), e.g. build_scene, add_text, set_property, add_camera_move — each call one undo step there.

The scene at the current time through the export path (what a PNG export produces), as an image.

無需輸入參數。

Export with any of the Export section's options, the files written on this computer as the tab renders them (no size limit): an image (png, jpg) at `seconds`, or a video (mp4, webm, gif, png-sequence) over `inSeconds`–`outSeconds`; `scale` 0.5–4, `fps`, `background: false` for transparency (PNG, WebM, PNG sequence); `artboard` an artboard's id or label, or "all" for one file each; `deliver` a destination's preset (app-store with `device` iphone/ipad/mac, keynote, powerpoint, google-slides). Only what you pass changes, and only for this export: the person's settings stay. `path` is a folder, or a file name for a single file (default ./paraplex-exports). A Free account gets the reason a Pro export is refused.

Render a variant per row, as Variants does: `rows` (objects) or `csv` (a header row, then a row per variant). Each column fills the layer named like it — its text, a counter's end value, a fill color, or a picture by its file name in the scene's media library — or the layer `bindings` maps it to (column → layer id or name). `nameColumn` names each file. Takes every option `export` takes (an image by default; kind video for a clip per row). Without rows, the sheet already set in Variants. Answers with the files and any column that bound to nothing.

Export the scene as a video with its own settings, written to a file in the export folder (or at `path`); answers with where it went. The `export` tool takes every option.

Import a HyperFrames project (HeyGen's HTML-to-video format) into the attached tab: `path` is its folder (index.html at the top) or a .zip of it. The whole project becomes one code layer — nested data-composition-src files, pictures, videos, fonts and stylesheets embedded, HyperFrames' runtime driven by the playhead — and each <audio> clip a sound layer at its start. Then look at it with frames, and finish it by hand or with run_tool.

Bring a file from this computer into the open scene, as dropping it on the canvas does: a picture or video becomes a layer (or fills the layer `target`, by id — a card, a phone screen, a component's slot), sound a clip at the playhead, a GLB/glTF/splat a model, a .lottie or Lottie .json an animation, an SVG a vector, a TTF/OTF/WOFF a font of the scene's. Answers with what happened and the layers it added.

Undo the last `steps` edits in the tab (default 1), as ⌘Z does — a run_tool call is one step. Answers with how many it took and whether more remain.

Redo `steps` undone edits (default 1), as ⇧⌘Z does.

A review sheet of the open scene, the way to judge a piece before calling it done: frames across it (default eight; or `times`) as one picture, and the flags — layers past the frame (more than 0.05 of the frame), text under 14 px on the frame, text under 3:1 contrast against what is behind it, motion hitches (a speed that jumps or a path that turns a corner mid-move). Fix what is flagged unless it is a deliberate choice you can name.

The piece on every review artboard (9:16, 4:5, 1:1, 16:9) as Responsive lays it out, at `times`: what each artboard did, what its frame cuts, the smallest type and type at the edge, with all the frames as one picture.

The person's projects, newest first: those in this browser and, when the tab is signed in, those kept in their account (in the cloud), each with its id, name, when it changed, and which one is open.

無需輸入參數。

Open one of the person's projects by id (list_projects), as the editor opens it: the scene being left is saved first; a project kept only in the account is fetched. Answers with its name and its scene.

Design systems the person can apply: Paraplex's and their own (made, imported or installed), each with what it holds (colors, text styles, materials, looks, fonts), and the one the scene follows now.

無需輸入參數。

Apply a design system by id or name, as Use in project does: its colors, text styles, materials, looks and fonts merged into the scene, which then follows it. One undo step. Then build with its tokens (run_tool get_brand reads them).

The person's skills and Paraplex's (Agent Skills: instructions and reference files), each with its description and whether it is on. Read one with load_skill when a task matches it.

無需輸入參數。

A skill's instructions (its SKILL.md body) and the files inside it, by name. Read it before doing the work it describes.

One file inside a skill — a reference, a template, a checklist — by the path load_skill listed.

Write a skill's SKILL.md into the person's library: the whole file, frontmatter (name, description) then instructions. Rewrites the named skill's SKILL.md and keeps its reference files, or creates it. Paraplex's own skills are not rewritten.

Save the open scene as a project of that browser's, by name.

The review's facts at a time: where every layer lands on the frame (NDC boxes, ±1 is the frame's edge), each text's size on the frame in px and its fill, the camera.