Claude & ChatGPT

What it is

Misu ships a connector built on MCP, the standard Claude and ChatGPT both use to talk to outside tools. Once connected, the chat can read your brands, products, trained characters and saved Looks, check your credit balance, generate stills and video, browse your Library, and make, fill and read moodboards. Renders land in that brand's Library, tagged as agent work.

It is the same engine as the Studio, not a lighter copy of it. Model routing, product references, Character Lock, the credit ledger and refunds all behave exactly as they do in the browser.

Connecting

Find the connector URL in Settings, then Connections. There is no API key — you sign in with the Misu account you already have, and the chat gets exactly the brands that account can reach.

In Claude: Settings, then Connectors, then Add custom connector. Name it Misu, paste the URL, sign in and approve. In ChatGPT: Settings, then Apps, then Advanced; switch on Developer mode, add an MCP server, paste the same URL.

What it costs

Reading is free — brands, products, characters, balance, job status and Library browsing never cost anything. Generating spends the brand's credits at Studio rates, so an image is typically 4 to 120 credits depending on model and quality, and a clip costs more. A failed render refunds itself.

Ask for a quote first and the chat gets one for free: credits for stills, dollars and the provider for video. A batch of up to eight stills is refused before anything is charged if it would cost more than 1,500 credits, so split a bigger set.

Permissions carry over. A member who cannot spend in the Studio cannot spend from a chat either.

Workflows and the results view

Six ready-made workflows ship with the connector: packshot set, colourways, Character Lock campaign, outfit to campaign, still to motion and a brand check. In Claude they sit in the prompt menu. In ChatGPT, ask the chat which Misu workflows there are. Each one plans the frames, quotes the cost, waits for your yes, renders and reports back in your own terms.

Where the chat app supports it, results open in a Misu panel inside the conversation: a gallery of the frames, a player for clips and a live status for anything still rendering. Variation, Animate and Use as reference write the request into the chat, so the cost is still quoted before anything is spent. Apps without the panel show the links and small previews as before.

Working from your own photo? The chat can ask for an upload link, so a local file becomes something Misu can render from.

Getting good results

The one habit that matters: let the chat pass product and character ids rather than describing them. An id renders the real product, because its reference photo and any trained lock travel with the request. A description renders something that resembles it.

We publish a playbook — one Markdown file covering the standing rules, every tool, what the connector cannot do, and templates for briefs that recur. Download it from Settings, then Connections, or from the MCP page, and drop it into a Claude Project so every session starts briefed.

What is not in it yet

The connector generates; it does not retouch. Editing tools, Wardrobe try-on, Campaigns, briefs and training are all Studio surfaces with no tool on the connector today. Moodboards can be made, filled, read and turned into a Look or a lookbook, but arranging one happens in Misu. Canvas techniques are a half case: a chat can list them and tell you which one fits, but running one happens in Canvas. Most of the generate controls you know from the Studio are here — Looks, style lock, draft mode, camera, image-to-image, first and last frame, reference clips and Motion Control — but the retouching half of the Studio is not.

See also