Hiroska is an MCP server as well as an app. Connect the assistant you already use and ask it for a video in plain words: it writes the film with the Director, generates the shots, builds the timeline and hands you the export, inside your account, and it tells you the cost before anything is spent.
A request in plain words becomes a finished video
The assistant has the same tools the buttons have. It opens projects, searches your library, writes a film with the Director, quotes the run, generates the shots, builds the timeline and exports. You talk; it works in your workspace, as you.
Plan, write and film: the Director writes the moments, the assistant quotes the run and starts it once you say yes
Build from what you have: it searches your library, looks at an asset before placing it, and lays out a timeline from a shot list
Generate single pieces too: images, video and audio from any model in the catalogue, priced before they are made
Export and hand over: the one thing it gives you a link to is the finished render
Sign in from the assistant, or paste a key
claude.ai, ChatGPT and Claude Desktop add Hiroska as a connector and send you here to approve it, the way you would approve a calendar app. Claude Code, Cursor, Windsurf and anything with a config file take an API key you make in Settings, shown once with the exact line each client needs.
Sign in: add a connector with the address hiroska.com/mcp and no key, then approve it on a screen that names the app and what it may do
API key: name it, choose full access or read only, pick an expiry, and copy the command or the JSON for your client
Every connection is listed under Settings → Connected AI, with when it was last active and what it has spent
Works on every plan, including Free; the tokens are the same tokens the buttons spend
A limit, a scope, and a quote before every spend
An assistant in a loop spends faster than a person notices, so the walls are built in. It is told to quote before anything that costs tokens, a read-only key cannot change or spend anything, and all your assistants together are capped per day at a number you choose.
A daily spending limit for everything connected: 1,000 tokens by default, any number you like, or none
Read only for an assistant that answers questions about your work; full access for one that makes things
Only your active workspace: it cannot see another workspace, your password or your billing
Disconnect an app or revoke a key and its next request fails at once
Questions
What is MCP, and do I need to understand it?
MCP (Model Context Protocol) is the open standard AI assistants use to work inside other apps, the same way Claude or ChatGPT connect to a calendar or a drive. You do not need to understand it. Add Hiroska as a connector, or paste a key, and the assistant finds the tools by itself.
Which assistants can I connect?
Anything that speaks MCP. claude.ai on the web and on mobile, Claude Desktop and ChatGPT sign in through the connector screen; Claude Code, Cursor, Windsurf and other clients with a config file take an API key. Nothing on our side is specific to one of them.
Does it cost extra?
No. Connecting is free on every plan, including Free. Generation costs the same tokens it costs you in the app, and reading, planning, checking a plan and exporting cost nothing. The assistant is told to quote before anything that spends, and your daily limit caps the total.
Can it spend all my tokens?
Not past the limit. By default all your connected assistants together may spend 1,000 tokens in 24 hours. Raise it, lower it or remove it under Settings → Connected AI, which also shows the last 24 hours and each key’s 30-day total. A film run checks its ceiling against what is left before it starts.
What can it not do?
Upload files from your computer: MCP has no file transfer, so footage on your disk goes in through the app and the assistant then finds it by name. It also cannot change your account, team or billing, see your password, or reach any workspace but the one you have active.
Is what it makes private?
It works inside your workspace only, and everything it generates lands in your library like your own uploads. What it reads to do the job (project names, scripts, a downscaled still when it asks to look at an asset) goes to the assistant’s provider the way your own messages do, so connect assistants you already trust with your work. Disconnect one and its access ends at once.
Can I keep editing while it works?
Yes. Keep a project open in a tab and the assistant’s changes arrive on reload. If it changes a project you are editing at the same moment, the app asks you to reload rather than overwrite either side, and anything it generates appears in your library as it finishes.