MCP App Store
by KeenEthics
RTST AI icon

RTST AI MCP Server for ChatGPT

By RTST AI (VISATECH S.R.L.)

Overview

RTST AI connects your rtst.ai account so you can create images and videos just by describing them. Finished images, and playable videos, appear right in the conversation. What you can do: - Generate images: text-to-image, editing your own photos, variations, grids and multiple samples, with each model's own sizes and aspect ratios. - Generate videos: text-to-video and image-to-video with first/last-frame control, reference images and optional audio. - Motion transfer: animate a character image with the motion from a reference video. - Use your own images: attach a photo, pick one in the in-chat uploader, or import any public image by URL. - Browse and reuse: your image, video and motion-transfer library, with one-click animate, variation, remix and use-as-reference. Generations are private to your rtst.ai library unless you ask to publish them. Sign in with your rtst.ai account, or create one during sign-in; your password is never shared and you can disconnect at any time.

How to connect

  1. ChatGPT: Open the RTST AI listing in the ChatGPT app directory and choose Connect. ChatGPT listing RTST AI exposes 18 tools here.

Tools

balance

ChatGPT
Return the current rtst credit balance for the authenticated user.

card_refresh

ChatGPT
Internal: used by RTST cards to refresh their own data. Do not call this — use job_status, list_images, list_videos or transactions instead.

enhance_prompt

ChatGPT
Rewrite/expand a short prompt into a richer generation prompt (LLM-assisted). Returns the enhanced prompt text; does not generate media. Feed the result to generate_image / generate_video.

estimate_cost

ChatGPT
Preflight the exact credit cost of an image or video job before submitting — pass the same arguments you will pass to generate_image / generate_video (model, samples, grid, size, reference images, params; or duration, resolution, aspect_ratio, audio, frames, reference_images). Returns credits, whether discounted, current_balance and can_afford. type "motion" (no model needed) returns the motion model's credits_per_second and, given the reference clip's duration, an estimate of rate × seconds (capped at max_video_duration).

generate_image

ChatGPT
Submit a text-to-image or image-to-image job. Returns generated image URLs directly when the job completes synchronously, otherwise a job_id to poll with job_status (kind:"image"). Reference images must be hosted https:// URLs — get one via media_upload (file for a ChatGPT attachment, base64 elsewhere), media_upload_widget (Apps-UI file picker), or media_import_url (web URL); raw local paths in this field are not supported. Output is PRIVATE to the user's library unless they ask to publish it (private:false). In ChatGPT the result renders as an inline card, so do not repeat the image URLs — summarize and suggest a next step.

generate_video

ChatGPT
Submit a text-to-video or image-to-video job. Always asynchronous: returns a job_id — poll job_status (kind:"video") until state is "success" (video typically takes 60-180s), then read the video URL. Video ALWAYS costs credits, and the price depends on resolution, length and audio, so the listed per-second rate is not a quote — run estimate_cost first and tell the user that number; never call a video job "free". Frame/reference images must be hosted https:// URLs (media_upload — file for a ChatGPT attachment, base64 elsewhere — media_upload_widget, or media_import_url). The video is PRIVATE to the user's library unless they ask to publish it (private:false). For reference_images, cite each one in the prompt as {{img1}}, {{img2}}… (in array order) so the model knows what each image is for. Seedance 2.x models take EITHER first/last frames OR reference_images in one job, not both. In ChatGPT the returned card polls the job itself and plays the video inline when done — you may still call job_status to report completion.

get_model

ChatGPT
Look up the ONE model the user named (codename or display name, e.g. "Nano Banana Pro", "seedance 2.5", "kling3") and return its exact name plus the same entry list_models gives: description, aspect ratios, durations, resolutions, audio/image support, pricing, and for image models the params they accept. Whenever the user has already specified a model, call this instead of list_models, then generate with the returned name and only the settings it lists. Several close matches come back as matches (ask the user or pick the one they meant); no match returns the available names.

job_status

ChatGPT
Poll an image, video, or motion-transfer job. Returns state (pending|success|failure|unknown), poll_after_seconds while pending, and result URLs when complete. Set sync:true to have the server wait up to ~25s for a terminal state before returning. In ChatGPT the generation card already follows the job: call this there only when the user asks about a job.

list_images

ChatGPT
List the authenticated user's generated images, most recent first (paginated). Each item has id, prompt, model, url, thumb, width/height, created_at and web_url. In ChatGPT the result renders as a gallery card — do not list every URL in prose.

list_models

ChatGPT
Browse the model catalog: available generation models with a per-model description (what each is best for) plus capabilities (aspect ratios, durations/resolutions, audio/image support, credit rates, is_free; image models also list their params — the controls they accept). Read description to pick the best model for the user's goal, then take aspect_ratio/duration/resolution (and, for images, any params key) only from that model. A video model's credits_per_second is only a base rate: resolution and audio move the real price either way (wan22 lists 5/s, yet 5 s at 480p costs 10 and at 1080p 30), so never quote a price worked out from it; models with pricing (Seedance 2.x) are token-priced and have no rate at all. Video is never free unless is_free is true; always call estimate_cost for the exact charge before generating and tell the user. Returns data only (no card). Use it to choose a model when the user has not named one: pick by description and price, then tell the user which model you chose and why. If the user named a model, call get_model instead. To show the user the models, call show_models. Motion transfer has no model choice (motion_transfer uses a fixed model).

list_videos

ChatGPT
List the authenticated user's video jobs, most recent first (paginated). kind "video" (default) is generate_video output; kind "motion" is motion_transfer output. Each item has id, prompt, model, status, url, thumb, duration, created_at and web_url. In ChatGPT the result renders as a gallery card — do not list every URL in prose.

media_import_url

ChatGPT
Fetch a public https:// image and re-host it in rtst storage, returning a hosted URL you can pass to generate_image (image) or generate_video (first_frame / last_frame / reference_images). Use this for remote image inputs — the server cannot read local files.

media_upload

ChatGPT
Upload an image and get a hosted https URL to use as a reference / frame / character input for generate_image, generate_video, or motion_transfer. In ChatGPT, when the user has ATTACHED an image to the conversation, pass it as file — no other step needed. Otherwise pass data_base64 (base64 of the file, or a full data:image/...;base64,... URI) — e.g. a local file read in Codex or Claude Code. In Apps-UI clients with no attachment (ChatGPT, claude.ai) use media_upload_widget so the user can pick a file. For an image already on the web, use media_import_url instead. Images only; max 15MB.

media_upload_widget

ChatGPT
Open an in-chat file picker to upload a LOCAL image and get a hosted URL for generate_image / generate_video / motion_transfer. Use this in Apps-UI clients (ChatGPT, claude.ai web/desktop) whenever the user wants to use their own photo and none is attached: call it right away in that turn, before choosing a model or quoting, instead of asking them in text to upload or attach it. In other clients use media_upload with data_base64. After calling it, wait for the user to pick a file — the widget uploads it and reports the hosted URL; do not also call media_upload.

motion_transfer

ChatGPT
Animate a character image with the motion from a reference video. Asynchronous: returns a job_id — poll job_status (kind:"motion") until state is "success". Both inputs are hosted https:// URLs (media_upload_widget in Apps-UI clients, media_upload base64 elsewhere, media_import_url for web URLs). Provide reference_video_url OR template_id. The reference video is one the user chose (their pick from list_videos, or a URL they gave): never pick it yourself. The model is fixed (Wan Motion Transfer): never ask the user to choose one. Quote with estimate_cost type "motion" first.

show_models

ChatGPT
Show the user the RTST models: in ChatGPT a picker card with one label per model and its price; clicking one sends a message naming that model. Call this ONLY when the user asks to see, browse or compare models, and once per request. Returns the same catalog as list_models. To choose a model yourself, use list_models (no card); for a model the user named, get_model.

transactions

ChatGPT
List the authenticated user's credit transactions (spend / earn / adjustments), most recent first (paginated). For the current balance use balance.

whoami

ChatGPT
Return the identity of the rtst account this connection is authenticated as — a stable account id, email, name, plan and credit balance. Use it to confirm which account is logged in (e.g. before spending credits, or when the user asks "which account is this?"). The account is derived from the connection's token; it cannot look up other users.

Capabilities

InteractiveWrites

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/rtst-ai"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/rtst-ai)