For creators
Give the agent you already use the ability to make images, video and audio. Nothing to integrate, no key to paste — one command and it can generate.
Connect it
Sequel speaks MCP, so it works in Claude Code, Cursor, ChatGPT and anything else that does. Pick your client and run the line it gives you.
There is no API key in that command. Your client opens a browser, you sign in and approve it there, and the token goes to the client rather than through your clipboard.
Teach it to use the tools well
An agent left to itself reaches for the model it has read about most. Paste this in so it searches first, checks the schema, and spends your money deliberately.
You have Sequel connected over MCP, giving you image, video and audio generation. Use the tools in this order: 1. search_models — describe the job, not the model. Prefer cheap and fast unless quality is asked for. 2. get_model_schemas — when you need a model's exact parameter names. 3. run_models — generate. Rules: - Never invent a model id. Only ids returned by search_models exist. - The input object is the model's own: prompt for most, text for speech. - Asked for several things, send ONE run_models call with a prompts array rather than one call each. They run in parallel. - Video is asynchronous: run_models returns in_progress with a task id. Say so rather than pretending it is finished. - You cannot see what you generated. Never describe its contents. - Write the prompt as prose, not comma-separated tags. Be concrete about materials, lens and lighting; quality tokens like "8K" do nothing. Say what should be in frame, never what should not.
Then just ask
Describe the job rather than the model — “four product shots on seamless white, warm side light”, “six seconds of the lamp flickering on”. The agent picks what fits the brief and the budget, and every run comes back with its cost attached.