
Script + screenshots in, a narrated product walkthrough out. Ships as a static web page or MP4.
Script + screenshots in, a narrated product walkthrough out. Ships as a static web page or MP4. Claude Code / Codex skill.
·
Guides that talk. Script + screenshots in, a narrated product walkthrough out. Ships as a small static web page (the sample is 1.4 MB), or as MP4 when you need a file.
A Claude Code / Codex skill. MIT.
Play the sample guide with sound (1:29, English). It walks through mermandraw.com: write the code, pick a theme, export, send a link. Everything that made it is in guides/first-diagram: a 197-word script, four captures, and one scene file where every highlight is cue(scene, word). Nothing was placed on a timeline by hand. The same guide in Korean, with a Korean diagram and voice: Korean sample (1:38, guides/first-diagram-ko). Both samples are narrated with ElevenLabs voices (Max in English, Anna Kim in Korean, eleven_v3). Without a key, the same guides build with the default Edge voices; the voice is set in tts in guide.json.
Every motion it can make plays live in the gallery.
You write the script. Each paragraph names the scene it belongs to and says what should be on screen. The agent captures your web app with a headless browser (or uses screenshots you bring), synthesizes the narration, and assembles the guide. Every motion on screen is cued to a word of the narration.
script.md + your web app (or screenshots)
│
▼
narration (TTS, word timestamps) → captions + scene timetable
│
▼
scenes: cue("install", "click Save") → spotlight, pin, zoom, cursor
│
├──▶ dist/<name>/index.html static page, searchable transcript, follow-along mode
└──▶ dist/<name>.mp4 + .srt when you need a file (YouTube, Slack, LMS)Product UI changes. A recorded video dies with it. A guide built from a script does not:
cue(scene, word) anchors every highlight, zoom and cursor move to the narration. Change a sentence, re-synthesize, and the screen follows.<iframe>.The quickest way is to hand your agent the link: "Install this skill: https://github.com/SilverNine/script-to-guide". Or pick one:
# Claude Code, Codex and other agents that read skills, via skills.sh
npx skills add SilverNine/script-to-guide
# Claude Code plugin
claude plugin marketplace add SilverNine/script-to-guide
claude plugin install script-to-guide@silvernine
# Codex (or Claude Code without the plugin system)
git clone https://github.com/SilverNine/script-to-guide.git ~/.script-to-guide
mkdir -p ~/.agents/skills ~/.claude/skills
ln -s ~/.script-to-guide ~/.agents/skills/script-to-guide
ln -s ~/.script-to-guide ~/.claude/skills/script-to-guideThen ask your agent for a guide ("make a 2 minute guide for our signup flow"). On first use it runs setup, which installs the engine's Node packages, Playwright's Chromium (about 150 MB) and a Python venv inside the skill folder. The first narration check downloads a whisper model (about 490 MB for English, 570 MB for other languages).
Codex reads user skills from ~/.agents/skills and follows the link (Codex docs). Its default sandbox has no network and asks before writing outside your project, so expect approval prompts for setup, screen capture and narration, and for setup writing into the skill folder.
You need macOS or Linux, Node 18+, Python 3.10+ (on Debian or Ubuntu also python3-venv) and ffmpeg. Windows is untested.
The skill folder is the engine only. Your guides live in your project under guides/<name>/ and builds go to dist/, so updating the skill never touches your work. Add these to your project's .gitignore:
dist/
guides/*/takes/Every step is one command. Point stg at the engine: ~/.script-to-guide for a clone, or ~/.claude/plugins/cache/silvernine/script-to-guide/<version> for the plugin.
stg() { node ~/.script-to-guide/scripts/stg.mjs "$@"; }
stg setup # once, and after an update
stg doctor # what is installed and what is missing
stg new signup # ./guides/signup/ from the sample, builds as is
stg synth signup # narration, captions, scene timetable
stg verify signup # transcribe it back and compare with the script
stg check signup # type-check the scenes
stg build signup # dist/signup/index.html
stg frames signup --at 0,450 # review stills
stg mp4 signup # dist/signup.mp4 + .srtdist/<name>/ is a static folder: index.html, one app.js, the captures and the narration. Upload it to any static host, then:
<iframe src="https://docs.example.com/guides/signup/" title="Signup guide"
style="width:100%;aspect-ratio:16/10;border:0" allow="fullscreen; clipboard-write"></iframe>| Layer | What it is |
|---|---|
SKILL.md | The skill entry point. Picks a mode, says which reference to read, states the rules the agent must not break |
references/ | Pipeline stages and checkpoints, script format, narration rules, pronunciation dictionary, capture, final review, build and MP4 |
cards/ | 13 motion cards: spotlight, numbered pin, magnifier, cursor click, callout, type-in, screen swap, camera push-in, redact, title, chapter, step list, drawn dialog. Each has an intent, when to use and avoid it, a parameter table, known pitfalls and an example |
gallery/ | Every card playing live, rendered by the player's own runtime |
player/ | The player: a single-page React app bundled into one app.js. The clock is the audio position, so picture and sound cannot drift. Keyboard shortcuts, captions, follow-along mode, light and dark themes |
scripts/ | stg.mjs, the one entry point: setup, new guide, capture, TTS with word timestamps, transcription check, web build, MP4 render |
guides/first-diagram/ | One complete sample guide for a public app: script, pronunciation dictionary, captures with element boxes, scenes |
guides/first-diagram-ko/ | The same guide in Korean: Korean script with English UI names, a Korean pronunciation dictionary, Korean-diagram captures |
Narration comes from a text-to-speech provider that returns word timestamps, so captions are the script word for word.
ELEVENLABS_API_KEY. The scripts never print the key. The sample guides use it..venv/bin/python -m pip install supertonic; the model (about 385 MB) downloads on first use and is licensed under BigScience Open RAIL-M, which restricts some uses. It returns no word timings, so words inside a sentence are placed by letter count and a highlight can land about 0.2 s early or late.The code is MIT. The voice output follows the terms of the provider you choose.
MIT. See LICENSE.
Paste this prompt into your agent. It reads this page and does the setup for you.
Read https://aiagentslisting.com/mcp/script-to-guide to learn what the "Script To Guide" MCP server does and how to install it. Add it to my coding agent's MCP configuration as documented on that page, then confirm the server connects and list the tools it exposes.Agents can also browse this directory over MCP at https://aiagentslisting.com/api/mcp
npx skills add SilverNine/script-to-guideThis server runs locally, so we can't read its tool list over the web yet.
We could not run this server: it ships no npx package for our sandbox and no public endpoint. The four checks that need a live server are left out, and the grade tops out at B until one exists.
Maintain this server? Claiming proves you own the listing and earns the last 4 points.
Claim this listingPaste this line near the top of your repository's README. The badge always renders the current score.
[](https://aiagentslisting.com/mcp/script-to-guide)Nothing comparable is listed yet.
v0.1.0v0.1.0: Guides that talkRelease notesConnect to the AI Agents Listing MCP server at https://aiagentslisting.com/api/mcp using the HTTP transport and complete the sign-in when it asks. Then call the submit_review tool for the listing "script-to-guide" with a rating from 1 to 5 and a short note on what worked and what did not.No reviews yet
Be the first to share your experience with this tool.
One email a week. New agents, MCP servers and skills, and what is actually getting traction.