Demo video pipeline for fleet repos. Records narrated walkthroughs of any webapp by orchestrating Playwright, speech-mcp TTS, and FFmpeg composition.
uvx mcpb install sandraschi/demo-vid-mcp- Generate a video:
demo_vid_generate(repo="chitchat")— auto-starts the target webapp, records it, adds voiceover, composes MP4 - Draft a script:
demo_vid_script_draft(repo="chitchat")— reads the README and generates a narration script - Browse the depot: Categorized gallery of produced videos with inline player, rebuild, delete, and insert into repo README
- Chat about it: Built-in chat with personalities, example prompts, and local LLM integration (Ollama/LM Studio)
| Tool | Description |
|---|---|
demo_vid_generate(repo, theme="dark") |
Full pipeline: auto-start target webapp → record (Playwright) → voiceover (speech-mcp) → compose (FFmpeg) → MP4. theme="light" forces the target webapp's light mode for bright demos (default "dark" matches fleet identity). |
demo_vid_script_draft(repo) |
Generate a narration YAML from the target repo's README and webapp page structure |
demo_vid_script_validate(yaml) |
Validate a narration script's structure and timing |
demo_vid_list(repo?) |
List produced videos with metadata |
demo_vid_refine(name, feedback) |
NOT YET IMPLEMENTED — edit narration.yaml directly and re-run generate |
demo_vid_help |
List all tools and usage |
| Page | Purpose |
|---|---|
| Dashboard | Backend status, KPI cards, pipeline overview |
| Depot | Categorized gallery with inline player, rebuild, delete, insert buttons |
| Generate | Select target repo by category → generate |
| Choreography | Visual script builder — 11 step types, global options, YAML preview |
| Chat | SOTA chat with personalities, localStorage, example prompts, LLM integration |
| Settings | LLM provider probe (Ollama/LM Studio), model selection, persistence |
| Logs | Ring-buffer log viewer with level filter and search |
| Help | 6-tab reference: overview, architecture, tools, config, fleet, troubleshooting |
demo_vid_generate automatically starts the target repo's backend and Vite frontend before recording. Zombie-kills stale processes first. No manual setup — the pipeline scans for start.ps1 and webapp/ directories.
Recording is dark-mode by default (fleet identity). Pass theme="light" to record a bright demo — the capture script forces the target webapp's theme class (handles both .dark toggling and persisted localStorage light-mode keys) before navigation, per chat_skills_prefab_standard.md §7.1.
Required: speech-mcp (voiceover), Playwright (recording), FFmpeg (composition). Optional: blender-mcp (3D titles), OBS-mcp (human-in-video), stems-mcp (music), vfx-mcp (effects).
Backend: 11134, Frontend: 11135, speech-mcp: 10909. See WEBAPP_PORTS.md.
| Doc | Contents |
|---|---|
| Installation | All install methods, prerequisites |
| Configuration | Env vars, config options |
| Tool Reference | All available tools |
| Development | Contributing, local setup |
| Troubleshooting | Common issues |
- Python 3.13+ with
uv - FFmpeg (for composition)
- speech-mcp (for voiceover)
- Playwright (for recording, auto-installed)
MIT