The AI that built your app knows every label on every screen. Give it the script guide, or connect it over MCP, and it writes narration that lands first time.
A tutorial script is a constrained form: one sentence per step, exact labels, open before you refer, a location clause on every button. Constrained forms are what language models are best at. The script guide states the rules with examples and tests; the MCP server lets the model check a draft against the same parser the extension uses, before you paste it.
From a single prompt to a closed loop.
The script guide ends with a prompt to paste into any assistant. Give it the labels and your demo values and it writes to the rules.
Connected over MCP, the AI can read the guides, parse a draft into beats, and get the same red flags rehearsal would raise.
After rehearsal, download a JSON of every beat that needed help with the fix for the script and the fix for the app. Hand it back to the AI.
Say what should change, by beat or by chapter, and the assistant rewrites, adds, or reorders beats to the rules.
Paste it into your project's instructions so new screens arrive with names, roles, and test ids.
The parser's rules are tested; the guide promises nothing the tests do not hold. Your AI writes to a contract.
The MCP server at tutituti.ca/api/mcp exposes the guides, a parser, and a checker. Claude Desktop, Claude Code, Cursor, and anything else that speaks MCP can connect in one line of config.
The script guide and its prompt, or connect over MCP.
Name the screens and the demo values. Get chapters and beats back.
Red flags become the rework file, which goes back to the AI. Two passes is typical.
Connect over MCP or paste the prompt.
Tuti Tuti - powered by Pandacat.