Skip to main content
MCP reference

The tools

About twenty actions. Your assistant picks the right one on its own: you never have to say these names out loud.

They're listed here for people who want to understand, and for the assistants that read this page before connecting.

Free​

These cost no cauris, and don't count towards the limits. That's deliberate: we want an assistant to check before it acts.

models_list​

The available models, with their starting cost.

Parameter
typeimage, video or audio. Optional.

Call this before any generation. The identifier it returns must be used character for character: the display name ("Seedance 2.0") is not an identifier.

account_balance​

Your balance in cauris and the last twenty transactions. No parameters.

account_profile​

The account profile: name, avatar and usage statistics. Not the balance: for cauris, use account_balance. No parameters.

generation_quote​

The exact price of a generation, in cauris, without launching or charging anything. The studio runs the same calculation as at launch: duration, resolution, sound, references, quality, count.

Parameter
typeimage, video or audio. Required.
model_idfrom models_list. Required.
promptthe planned prompt. Required.
the othersthe same as images_generate, video_generate or audio_generate

Your assistant states this price, waits for your go-ahead, then passes it as expected_cost at launch: if the price changed in the meantime, nothing is launched or charged, and the new price is returned.

References must already be online: the price of a reference video depends on its duration. It covers neither voice-over (audio_tts) nor sound effects: their prices are given further down.

voices_list​

The voices available for audio_tts: your own voices first, if you have any, then the standard voices. Returns the identifier to pass as voice_id. No parameters. Call it when you ask for a specific voice.

mes_medias​

The files already uploaded to the account (through televerser_media or from the website), with their address. Lets you reuse a reference without uploading it twice.

Parameter
typefilter by image, video or audio

televerser_media​

Uploads a file (image, video or audio) encoded in base64, and returns its public address, to use afterwards as a reference in a generation.

Parameter
content_base64the file content, in base64. Required.
file_namewith its extension, for example photo.jpg. Required.
mime_typeonly if the extension isn't enough to guess the type

Base64 is a third heavier than the original file and goes through the conversation. If the file is already online, prefer importer_media_url.

Limits: 10 MB for an image, 50 MB for audio, 100 MB for a video.

importer_media_url​

Downloads an image, a video or a sound from a public address, on the server side, and stores it in My media. Returns the final Pelikulart address, to use as a reference. Nothing goes through the conversation: this is the right method whenever the file is already online, or when its address is temporary and may expire.

Parameter
urlpublic http(s) address of the file. Required.
file_namename to give the file in My media. Taken from the address otherwise.

25 MB at most.

Past creations, newest first. Useful for finding an image you already made so you can reuse it as a reference.

Parameter
typefilter by image, video or audio
projetonly this project's creations (id or exact name)
cursorthe next page

projets_liste​

The account's open projects, as in My Creations in the studio: id, name, number of creations and cauris spent. A project groups creations together.

projet_creer​

Creates a project (50 open projects at most) and returns its id.

Parameter
nom24 characters at most. “Personnel” is reserved for the default project. Required.
descriptionone sentence about the project, 200 characters at most

ranger_creation​

Files existing creations in a project, or takes them out of their project.

Parameter
idsthe creations to file, 50 at most. Required.
projetproject id or exact name. Empty string: back to the default project, “Personnel”. Required.
An unknown name launches nothing

With the projet parameter, a generation whose project doesn't exist isn't launched and no cauris are spent: the tool says so and lists the account's projects.

creation_status​

Where a generation stands: queued, running, finished or failed.

Parameter
idthe identifier returned when it started. Required.

Leave at least ten seconds between two checks. Polling in a tight loop makes nothing faster.

upload_show​

Opens a window in the conversation where you drop your own files (images, videos, sounds) from your device. They go to My media in full quality, without passing through the assistant. This is the right choice whenever the file is on your computer or phone, if your assistant can display MCP app windows.

Parameter
typeauto (default), image, video or audio
multipleseveral files at once. Yes by default.

upload_request and upload_finalize​

For a large file (mostly video), when the assistant has a terminal. upload_request returns a signed upload address (upload_url), valid for 10 minutes. The file is sent straight to it with PUT, using the returned Content-Type header, without going through the conversation. upload_finalize then registers it and returns its permanent address.

upload_request parameter
file_namewith its extension, for example video.mp4. Required.
mime_typeonly if the extension is not enough
upload_finalize parameter
keyfrom upload_request. Required.
idfrom upload_request. Required.
file_nameoriginal name, shown in My media
mime_typeonly if the type sent differs

Without upload_finalize, the file stays invisible even though it arrived.

upload_start, upload_chunk and upload_complete​

Last resort, for a small file (1.5 MB at most) the assistant already holds in base64 and cannot send any other way. It announces the file with upload_start, sends it in pieces with upload_chunk, then upload_complete joins them, checks the size and returns the file address.

Parameter
file_nameupload_start: name with extension. Required.
total_sizeupload_start: exact file size in bytes. Required.
mime_typeupload_start: only if the extension is not enough
upload_idupload_chunk and upload_complete: from upload_start. Required.
indexupload_chunk: position of the piece, starting at 0. Required.
content_base64upload_chunk: this piece of the base64, 20,000 to 30,000 characters. Required.

Sending a piece again with the same index replaces it. If one is missing, upload_complete says which.

Which upload to choose

File already online: importer_media_url. File on your device: upload_show. Large file and an assistant with a terminal: upload_request. Small file the assistant already has: televerser_media.

The ones that cost cauris​

images_generate​

Starts an image. Returns an identifier: generation is asynchronous, so you then have to poll creation_status.

Parameter
model_idfrom models_list. Required.
promptthe description. Required.
aspect_ratio1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3. Square by default.
qualityif the model offers it
countnumber of images. Each one is charged.
image_urlsreference images
projetoptional: the project to file the creation in, by id or exact name (projets_liste)
Really describe what you want

A two-word prompt often gives you a failure or a forgettable image. Models respond far better to a full sentence: the subject, the light, the framing, the mood.

video_generate​

Starts a video. Expect several minutes.

Parameter
model_idfrom models_list. Required.
promptthe description. Required.
durationin seconds. Required: the price depends on it.
resolution720p, 1080p… Required: the price depends on it.
audiogenerate sound. Changes the price on some models.
image_urlsreference images
projetoptional: the project to file the creation in, by id or exact name (projets_liste)

Duration and resolution are required because the price is worked out from them. Without them, a fifteen-second video would be charged like a five-second one.

audio_tts​

Turns text into voice. Unlike images and videos, the result comes back immediately.

Parameter
text5,000 characters maximum. Required.
voice_ida specific voice, from voices_list. A default voice otherwise.
model_ideleven_v4 (default), eleven_v4_turbo (fast) or eleven_v3
projetoptional: the project to file the creation in, by id or exact name (projets_liste)

Price: 9 cauris per 500 characters (9 cauris up to 500, 18 up to 1,000, and so on up to 90 cauris for 5,000). With eleven_v4_turbo, 5 cauris per 500 characters (50 cauris for 5,000).

audio_generate​

Music and sound effects. The model chosen decides what it can do; a short effect and a full song are priced very differently, and models_list gives the prices.

Parameter
model_idfrom models_list. Required.
promptwhat you want to hear. Required.
projetoptional: the project to file the creation in, by id or exact name (projets_liste)

Deletion​

These two cost no cauris, but they are irreversible. A good assistant asks for confirmation before calling them.

supprimer_media​

Deletes an uploaded file. Not a generation: see supprimer_generation for that.

Parameter
idthe media identifier, from mes_medias. Required.

supprimer_generation​

Deletes a generation (image, video or audio created on Pelikulart). Not an uploaded file: see supprimer_media for that. No cauris are refunded: the generation has already used what it had to use.

Parameter
idthe generation identifier, from creations_search or creation_status. Required.

How a generation goes​

  1. Your assistant calls models_list to learn the identifiers
  2. It starts images_generate or video_generate and gets an identifier back
  3. Cauris are charged at the start
  4. It polls creation_status from time to time
  5. When it's ready, the address of the image or video is returned

If a generation fails, the cauris are refunded automatically. You don't have to ask.