MCP App Store
by KeenEthics
idea.ai icon

idea.ai MCP Server for ChatGPT

By idea.ai

Overview

Create and edit images, video clips, audio, and 3D assets with idea.ai. Twelve guided workflows help you maintain character consistency, create product campaigns, develop storyboards and video shots, transfer motion, make talking presenters, localize videos, adapt social layouts, produce audio, restore assets, develop 3D models, extract highlights, and create vector artwork concepts. Reuse reference assets and refine results through follow-up requests. Available models, controls, and output formats vary by workflow.

How to connect

  1. ChatGPT: Open the idea.ai listing in the ChatGPT app directory and choose Connect. ChatGPT listing idea.ai exposes 20 tools here.

Tools

check_generation

ChatGPT
Report a generation's status to the views that poll it — the upload picker and the generation viewer. This is their own tool and renders nothing — use get_generation, which shows the user their result.

create_3d_model

ChatGPT
Generates a 3D model from a text prompt with a text-to-3D model, or, when an image is passed in files, from that image with an image-to-3D model. It spends coins from the user's idea.ai balance and a completed generation is not refunded, so confirm the price with the user first using quote_cost with the same prompt, options and all priced files. Preserve the quoted settings and resolvedOptions; reference fitting can add to a standard model's base price, and reserve-and-settle models can hold or settle a different amount. Pick the model from list_models for this operation (modality textTo3D for a prompt alone, imageTo3D with input files); option names, allowed values and file roles are per model (get_model). Each file is a public url, a result url from get_generation, or an uploadId from upload_media or create_upload, with the role the model expects; a reference role needs an alias that the prompt cites as {{alias}}. Returns at once with a generationId; where the host shows the generation view, it displays the result. Replaces generate for textTo3D and imageTo3D.

create_audio

ChatGPT
Generates speech, dialogue, music or sound effects from a text prompt with a text-to-audio model; a speech model's voice id comes from list_voices. For structured dialogue models, supply the script and voices in options.inputs; a separate prompt is optional. When files are passed, produces audio from input audio or video with a media-to-audio model, such as cleaned-up speech, a voice separated from its background, or speech in a reference voice. It spends coins from the user's idea.ai balance and a completed generation is not refunded, so confirm the price with the user first using quote_cost with the same prompt, options and all priced files. Preserve the quoted settings and resolvedOptions; reference fitting can add to a standard model's base price, and reserve-and-settle models can hold or settle a different amount. Pick the model from list_models for this operation (modality textToAudio for a prompt alone, mediaToAudio with input files); option names, allowed values and file roles are per model (get_model). Each file is a public url, a result url from get_generation, or an uploadId from upload_media or create_upload, with the role the model expects; a reference role needs an alias that the prompt cites as {{alias}}. Returns at once with a generationId; where the host shows the generation view, it displays the result. Replaces generate for textToAudio and mediaToAudio.

create_image

ChatGPT
Generates images from a text prompt with a text-to-image model, or, when files are passed, from a source video, with reference images where the model takes them, with a media-to-image model. It spends coins from the user's idea.ai balance and a completed generation is not refunded, so confirm the price with the user first using quote_cost with the same prompt, options and all priced files. Preserve the quoted settings and resolvedOptions; reference fitting can add to a standard model's base price, and reserve-and-settle models can hold or settle a different amount. Pick the model from list_models for this operation (modality textToImage for a prompt alone, mediaToImage with input files); option names, allowed values and file roles are per model (get_model). Each file is a public url, a result url from get_generation, or an uploadId from upload_media or create_upload, with the role the model expects; a reference role needs an alias that the prompt cites as {{alias}}. Returns at once with a generationId; where the host shows the generation view, it displays the result. Replaces generate for textToImage and mediaToImage.

create_upload

ChatGPT
Creates a new upload session and a presigned PUT URL for one local file you can read yourself. This call transfers no file bytes, spends no coins and starts no generation. Each call reserves a fresh upload id and storage object key. For a batch slot, the newest session becomes the selection returned by get_upload_batch; the previous upload and stored file are not overwritten or deleted and remain subject to their normal expiry. PUT the raw bytes separately before uploadExpiresAt, with the returned mimeType as Content-Type and no Authorization header, then call get_upload until ready. Pass uploadId as files[].uploadId to a separate paid generation tool after quoting and confirming its cost. The file name's extension sets the type.

create_video

ChatGPT
Generates a video from a text prompt with a text-to-video model, or, when files are passed, generates or edits a video from input media — frames, reference images, audio or a source video, as the model takes them — with a media-to-video model. It spends coins from the user's idea.ai balance and a completed generation is not refunded, so confirm the price with the user first using quote_cost with the same prompt, options and all priced files. Preserve the quoted settings and resolvedOptions; reference fitting can add to a standard model's base price, and reserve-and-settle models can hold or settle a different amount. Pick the model from list_models for this operation (modality textToVideo for a prompt alone, mediaToVideo with input files); option names, allowed values and file roles are per model (get_model). Each file is a public url, a result url from get_generation, or an uploadId from upload_media or create_upload, with the role the model expects; a reference role needs an alias that the prompt cites as {{alias}}. Returns at once with a generationId; where the host shows the generation view, it displays the result. Replaces generate for textToVideo and mediaToVideo.

edit_image

ChatGPT
Edits or transforms one or more input images as a text prompt describes, with an image-to-image model. It spends coins from the user's idea.ai balance and a completed generation is not refunded, so confirm the price with the user first using quote_cost with the same prompt, options and all priced files. Preserve the quoted settings and resolvedOptions; reference fitting can add to a standard model's base price, and reserve-and-settle models can hold or settle a different amount. Pick the model from list_models for this operation (modality imageToImage); option names, allowed values and file roles are per model (get_model). Each file is a public url, a result url from get_generation, or an uploadId from upload_media or create_upload, with the role the model expects; a reference role needs an alias that the prompt cites as {{alias}}. Returns at once with a generationId; where the host shows the generation view, it displays the result. Replaces generate for imageToImage.

extract_text_from_media

ChatGPT
Extracts text from input media with a media-to-text model, such as a transcript of audio or video, or the contents of an image or document. It spends coins from the user's idea.ai balance and a completed generation is not refunded, so confirm the price with the user first using quote_cost with the same prompt, options and all priced files. Preserve the quoted settings and resolvedOptions; reference fitting can add to a standard model's base price, and reserve-and-settle models can hold or settle a different amount. Pick the model from list_models for this operation (modality mediaToText); option names, allowed values and file roles are per model (get_model). Each file is a public url, a result url from get_generation, or an uploadId from upload_media or create_upload, with the role the model expects; a reference role needs an alias that the prompt cites as {{alias}}. Returns at once with a generationId; where the host shows the generation view, it displays the result. Replaces generate for mediaToText.

generate

ChatGPT
[DEPRECATED, removed on 16 December 2026] Use create_image, edit_image, create_video, create_audio, extract_text_from_media or create_3d_model instead; each does one job with its own schema. This tool still works until then and spends coins exactly as before. It takes the same model, prompt, options and files plus a modality. A file declared by fileName and fileSizeBytes instead of a url still returns a presigned uploadUrl to PUT the raw bytes to, not multipart, until removal.

generate_with_upload

ChatGPT
[DEPRECATED, removed on 16 December 2026] Use upload_media to upload the user's files, then the create_, edit_ or extract_* tool for the operation. This tool still works until then and spends coins as before. It shows the user a file picker and starts the generation itself, charging coins, once every file is picked.

get_generation

ChatGPT
Show a generation's result, or look at an earlier one again. Free and read-only. A create_, edit_ or extract_* call already renders its own self-updating view, so do NOT call this right after one: it stacks a duplicate copy of that view in the conversation. Call it when the user asks about an earlier generation, when you need to see an image yourself to describe or iterate on it, or ONCE when a generation id arrives from a finished generate_with_upload picker, which shows no result itself. Once completed it returns the image inline as an image block — a downscaled preview; raise limit (max 4) to see more of a set — plus each output's full-resolution url, which works as files[].url on a follow-up edit, animation or extension. Where the view plays audio or video itself, no url for them appears in the text: do not dig one out of the structured output to paste into your reply; a url that does appear is one the view cannot play, so hand it to the user. A 3D model url is one of those: for a GLB or glTF file, call preview_3d once to open the interactive preview, and still hand the url over. A failed generation returns its reason as a normal result, not a tool error. elapsedSeconds is how long it has been running, or took; retryAfterSeconds is the view's own polling cadence, not an instruction to call this again.

get_model

ChatGPT
Get one model's full capabilities, keyed by modality under capabilities: inputSchema is a JSON Schema for that modality's options (types, enums, ranges, defaults), and files carries the input-media constraints — roles with the minimum and maximum number of files each accepts, whether a role needs an alias, its accepted formats and size, duration and dimension limits, plus cross-role rules such as mutually exclusive groups and combined totals. pricing gives the coin cost: a base coinAmount when its type is "standard"; when "dynamic", pricing.inputs lists the options and file metadata the price depends on — send exactly those to quote_cost for the amount. An aspectFit rule can add a surcharge per reference image even to standard pricing; send its option and every affected image's dimensions to quote_cost. An option left out of a create_, edit_ or extract_* call takes the default shown in inputSchema, resolution included, so read the defaults before comparing models. Option names, allowed values and file roles are all per-model, and a wrong one is rejected. voiceCatalog, when present, is the speaker catalog this model draws from: pass it to list_voices to find valid ids for its voice option, rather than guessing one.

get_upload

ChatGPT
Check an upload from upload_media or create_upload. Free and read-only. While the upload is pending it waits up to about 15 seconds for it to become ready, failed or expired. A ready upload carries the measured type, size, dimensions, duration and frame rate, and works until expiresAt as files[].uploadId on the create_, edit_ or extract_* tool and as inputMedia[].uploadId on quote_cost.

get_upload_batch

ChatGPT
Lists the uploads the picker made for one upload_media call, by its batchId. Free and read-only: call it when the user's next message after upload_media arrives, and again if any upload is still pending. Each upload comes with the role and alias the picker gave it and, once ready, its measured type, size, dimensions, duration and frame rate; a ready upload works until expiresAt as files[].uploadId on the create_, edit_ or extract_* tool and as inputMedia[].uploadId on quote_cost. A batch the user has not dropped files into yet, an unknown one and an expired one all list nothing.

list_models

ChatGPT
List the generation models available, optionally filtered to those supporting one modality — each create_, edit_ and extract_* tool names the modality it runs. Returns each model's id, display name, provider, the modalities it supports, its prompt-length limit where it has one, and its pricing — "standard" carries a fixed coinAmount, "dynamic" is priced per request and quote_cost gives its amount. A model's options and input-file constraints are not included; get_model has them. A voiceCatalog means the model speaks with a catalog voice, whose ids list_voices returns.

list_voices

ChatGPT
List or narrow the selectable speaker voices in one catalog. Pass a returned voice id as the model's voice option — speaker on Seed Audio, inputs[].voice.voice_id on ElevenLabs v3. Never invent a voice id; it will be rejected. Take the catalog id from voiceCatalog on get_model or list_models. A voice id is only valid on the catalog it came from, so a model backed by a different catalog rejects it. ASK BEFORE YOU LIST. A catalog holds hundreds of voices and "a voice" is a matter of taste, so unless the user has already said what they want, call with facetsOnly: true first. That returns no voices — only the gender, age, accent, useCase and language values the catalog actually has, each with a count. Catalogs differ in what they describe, so a dimension none of its voices carry is left out; do not offer it. Offer those to the user, then call again with their answer. Do not dump the catalog at them, and do not quietly pick for them. facetsOnly also narrows: send the filters agreed so far and it reports what is still available within that set, so you can ask one question at a time (gender, then accent) without fetching voices in between. Filters match genai's normalized lowercase labels: gender, age and useCase are exact, accent is a substring so "zealand" finds "new_zealand", language takes a lowercase ISO 639-1 code, region-qualified with a hyphen as the facets list it ("pt-br"), where a bare code also matches its regions ("pt" finds "pt-br"), and search runs over name and description. When nothing matches, the response carries the whole catalog's available values — read those and retry rather than reporting that no such voice exists. A dimension missing from them is one this catalog does not describe: drop that filter. Returns matched (how many passed the filter) and total (the catalog size) alongside the voices, capped at limit — default 30, max 200. Narrow with filters rather than raising limit to page through a whole catalog.

preview_3d

ChatGPT
Open the interactive 3D preview of a generation that already exists. Read-only: it never starts a generation and never spends coins. Call it once when a completed output is a GLB or glTF file, after wait_for_generation reports that output, or when the user asks to see the model. Other 3D formats (OBJ, FBX, STL, USDZ) do not preview: hand over their url instead. Do not start another generation just to preview. The view orbits and zooms and keeps Download and Open. If the preview cannot load, the model url in this result is still the file.

quote_cost

ChatGPT
Get the coin quote for one generation before running it. Free and read-only: nothing is generated or charged. Send the same prompt and options you intend to generate with and one inputMedia entry for every priced file, including every reference whose dimensions affect an aspectFit rule. A standard model's pricing.coinAmount is its base price: reference fitting can add a per-image surcharge, so quote reference-based requests too. Dynamic models require every option and media field listed in pricing.inputs. A ready upload's uploadId supplies its measured metadata; otherwise supply the required metadata yourself. Missing required dynamic pricing inputs are rejected rather than guessed. The result carries coinAmount and any resolvedOptions. Confirm the quote with the user and preserve the same model, prompt, complete file set, metadata and options; merge resolvedOptions into those options before generating. Changing any priced input needs a new quote. On reserve-and-settle models the reservation and final charge can differ: Seed Audio quotes a maximum for generated audio; Clean Audio settles on vendor-measured input duration and, for URL-sourced audio, reserves the role maximum even when the quote is lower. Explain these conditions before spending.

upload_media

ChatGPT
Shows the user a picker to upload files from their own device to idea.ai — files they would pick off their machine, paste or drag in, when you have no url for them. Spends nothing. Not for a result this server produced: that already has a url, which goes in files[].url. The user's browser uploads the bytes, so your own network access does not matter: never ask the user to host the file or to change a network or egress setting. Returns before the user picks, with a batchId: once the user says the files are in, get_upload_batch with that batchId returns each upload's id, role, alias and measured type, size and duration. Use the ids as files[].uploadId on the create_, edit_ or extract_* tool, after quoting the cost with quote_cost and confirming it with the user.

wait_for_generation

ChatGPT
Wait for a running generation to finish, for up to about 45 seconds per call, and return its status. Free and read-only, and it renders nothing — so unlike get_generation it never adds a copy of the view to the conversation. After a create_, edit_ or extract_* call, call it and keep calling it with the same id until the generation finishes, then tell the user it is ready: your reply then lands with the result, which is what lets the host notify a user who stepped away. Also use it when you need the outcome to carry on yourself: comparing two results, or feeding one into the next generation. Each call holds for 45 seconds at most. Video takes a few minutes, up to an hour on some models. Once you have waited about ten minutes in total, end your turn and tell the user the view will show it when it is done, unless they asked you to wait it out. A result's elapsed time counts from creation, so after an upload it starts high: go by your own waiting, not by it. A completed result gives each output's width and height, the options the run used, and its full-resolution url for reuse as files[].url on a follow-up. It does not inline images or render a view — if the user has no view of this generation yet, call get_generation once to show it.

Capabilities

WritesInteractive

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/idea-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/idea-ai)