When it doesn't work
The cases below are ones we've actually run into, not a theoretical list.
The assistant can't see Pelikulart
Almost always the cause: your tool hasn't re-read its configuration.
Most clients only read their settings file at startup. Closing the window isn't enough.
Quit the app completely (Cmd + Q on Mac) then open it again.
This is the mistake that wastes the most time, by far. An assistant that's missing the server doesn't say "I'm missing the server": it improvises, opens random pages, or explains how to create a key that doesn't exist.
It's asking me for a key
There isn't one. Pelikulart connects by button only.
If your assistant insists on a token, it hasn't seen the server yet: see the point above. Once connected, it gets a polite refusal telling it exactly where to send you, and it opens your browser on its own.
The browser doesn't open
Some clients can't read the sign-in signal. In that case they can't connect to Pelikulart.
For Claude Code, start the sign-in by hand:
/mcp
Pick pelikulart, then Authenticate.
For Codex:
codex mcp login pelikulart
"Authorisation required"
Your authorisation was cut off, or it expired.
Reconnect from your assistant. If you cut off access from the MCP page, this is normal: you have to authorise again.
"Unknown model"
The identifier has to come from models_list, character for character.
Display names ("Seedance 2.0", "Flux 2 Pro") are not identifiers. Ask
for the list, then take the value of the id field.
Good news: an unknown model is turned down before anything is charged. It costs nothing.
"Not enough cauris"
Your balance is empty. Retrying won't make cauris appear.
"Too many requests"
One of the three limits has been reached. Wait the time given.
See The limits.
The generation failed
The cauris are refunded automatically. You don't have to ask.
Two common causes:
A prompt that's too short. Two or three words are rarely enough. Some models then answer with text instead of an image, and the generation is counted as failed. Really describe what you want.
The provider was busy. It happens. Try once more.
If the same model fails three times in a row, try another one and let us know.
It's stuck on "running"
An image takes a few seconds, a video often several minutes, sometimes longer at busy hours.
You can close the conversation: the generation carries on, and the result shows up in your gallery.
Something else
Write to support@pelikulart.ai with the name of your tool and what you saw on screen.