MCP: connect your AI to Tuti Tuti

Tuti Tuti runs a Model Context Protocol server at https://tutituti.ca/api/mcp. Connect Claude, Cursor, or any MCP client and it can read the guides, parse a draft script with the extension’s own rules, and get the warnings rehearsal would raise, before you paste a word into the panel.

Connect in one line

Claude Code

claude mcp add --transport http tuti-tuti https://tutituti.ca/api/mcp

Claude Desktop

Settings → Developer → Edit config, then add:

{
  "mcpServers": {
    "tuti-tuti": {
      "url": "https://tutituti.ca/api/mcp"
    }
  }
}

Cursor, Windsurf, and others

In the client’s MCP settings, add a server with the URL above. Cursor’s .cursor/mcp.json looks like this:

{
  "mcpServers": {
    "tuti-tuti": {
      "url": "https://tutituti.ca/api/mcp"
    }
  }
}

No key, no account, no session. The transport is streamable HTTP, stateless.

Tools

tuti_script_guide

The whole script guide: the rules, the examples, the prompt. Read before writing or rewriting a script.

tuti_app_guide

The app guide: what to build into the app so scripts land first time, with the paragraph to paste into a coding assistant's instructions.

tuti_example_script

A complete tour in the exact format, the same file the parser's tests hold to.

tuti_parse_script

Parses a script with the extension's own rules and returns chapters and beats with kind, action, target, typed value, and trigger word.

tuti_check_script

Checks a script against the guide and returns warnings per beat with the fix: missing transitions, pronoun targets, clicks without a location, values not in quotes, menus referred to before being opened, symbols the voice reads badly.

Resources tuti://guide/script, tuti://guide/app, and tuti://example/quick-tour expose the same documents to clients that prefer resources.

The loop your AI should run

  1. Call tuti_script_guide and, if it is building the app, tuti_app_guide.
  2. Write the script: one sentence per beat, exact on-screen labels, chapters separated by blank lines with a short title line.
  3. Call tuti_check_script and fix every warning. Repeat until it returns none.
  4. Call tuti_parse_script and confirm each sentence has the kind and action intended.
  5. Hand you the script. You paste it, rehearse, and send back the rework file if anything still misses.

A one-line instruction to your assistant covers it: “Use the tuti-tuti MCP server. Read the script guide, write the script for these screens with these demo values, and check it until there are no warnings.”

Try it from a terminal

curl -s https://tutituti.ca/api/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

MCP: the questions people ask

What is the Tuti Tuti MCP server for?
So the AI that writes your tutorial script can read the rules and check its draft against the same parser the extension uses, before you paste it. Fewer red flags in rehearsal, fewer rounds.
Which clients work?
Anything that speaks MCP over streamable HTTP: Claude Desktop, Claude Code, Cursor, Windsurf, Zed, Continue, and custom agents using the official SDKs. The endpoint is stateless, so no session handling is needed.
Do I need an account or a key to use it?
No. The guides are public and the checker holds nothing. If a future tool needs your account, such as reading your rework files, it will say so and use a token.
Is anything I send stored?
No. Scripts sent to the parser or checker are processed in memory and returned. There is no logging of script content.
Can my AI write the whole script through MCP?
It writes the script itself; the server gives it the rules and the checker. The loop is: read the guide, write, check, fix warnings, parse to confirm, hand you the text to paste into the panel.
Will there be write access, like saving scripts to my library?
Planned. Tutorials live in each person's browser today; a signed-in MCP tool that reads your rework file and writes a rewritten script back is on the roadmap.

Tuti Tuti - powered by Pandacat.

See also: AI-written scripts and Tuti Tuti for developers.