Sandbox
@mcpware/pagecast

MCP server for browser demo recording

Pagecast lets an agent open a real browser, interact with your app, and save the session as a demo GIF or MP4. It records clicks, typing, hover, and scroll events, then uses ffmpeg to add zoom overlays or cinematic camera movement before export.

47 stars5 forksJavaScriptUpdated 5mo ago
Who it's for

Builders who want their agent to record app demos, README GIFs, or social videos from a live browser.

What it delivers

You can turn a browser session into a shipping-ready demo file without manual recording or video editing.

What it does

Browser recording

Opens Chromium, records page interactions, and saves a WebM session with a timeline JSON file.

Interaction capture

Tracks click, type, hover, scroll, key press, and navigation actions with bounding boxes and timestamps.

Tooltip export

Creates a magnified inset near each interaction so small UI changes are easier to see.

Cinematic export

Crops and pans the camera to follow actions across the recording.

GIF and MP4 conversion

Converts WebM recordings into optimized GIF or MP4 outputs with ffmpeg.

Platform presets

Chooses sizes and formats for README, YouTube, TikTok, Reels, LinkedIn, and custom targets.

MCP tool set

Provides `record_page`, `interact_page`, `stop_recording`, `smart_export`, `cinematic_export`, `convert_to_gif`, `convert_to_mp4`, `record_and_export`, and `list_recordings`.

How to get it

  1. 1With Pagecast, you tell your AI
    "Record a demo of localhost:3000 for my GitHub README"
  2. 2Don't need zoom effects? Pagecast works as a plain screen recorder too
    "Record my app for 10 seconds and export as MP4"
  3. 3You don't need to know viewport sizes or formats. Just tell your AI the destination
    "Record a demo for GitHub README"           → 1280×720 GIF
    "Record my app for Instagram Reels"         → 1080×1920 MP4
    "Make a TikTok demo of my dashboard"        → 1080×1920 MP4
    "Record for YouTube"                        → 1280×720 MP4

README

Pagecast

npm version npm downloads license GitHub stars GitHub forks

English | 廣東話

Turn AI browser interactions into polished product demos.

Tell your AI to demo your app. Pagecast records the browser, tracks every click and keystroke, and exports a shipping-ready GIF or MP4 — with tooltip zoom overlays and cinematic pan effects. No screen recorder. No video editor. No post-production. Make a demo gif automatically after every PR if you want.

Without Pagecast — plain screen recording, you do the demonstration yourself and record yourself and you need to repead every time you changed the UI. You can see the cursor moving, but the actual interactions are too small to follow:

Original

With Pagecast (tooltip mode) — a magnified close-up appears on every interaction so viewers can actually see what's happening, and it design what to demo and make a new demo gif automatically:

Tooltip Demo

With Pagecast (cinematic mode) — the camera crops and pans to follow each action, and it design what to demo and make a new demo gif automatically:

Cinematic Demo

Two ways to use Pagecast

1. Product demo tool (the main use case)

You built a web app. You need a demo GIF for the README. Normally you'd:

  1. Open a screen recorder, manually click through the demo
  2. Open a video editor, zoom into the important parts
  3. Export, figure out ffmpeg, optimize the file size
  4. UI changes → repeat everything

With Pagecast, you tell your AI:

"Record a demo of localhost:3000 for my GitHub README"

The AI opens a real browser, interacts with your app, and exports a polished GIF with auto-zoom on every interaction. UI changes? Just re-run. The demo rebuilds itself.

2. Screen recording tool (also works)

Don't need zoom effects? Pagecast works as a plain screen recorder too:

"Record my app for 10 seconds and export as MP4"

The AI records, you get a .webm → optimized .gif or .mp4. Two-pass ffmpeg palette optimization, platform-aware sizing, one tool call.

Export Modes

ModeToolWhat it doesBest for
Tooltipsmart_exportFull viewport visible + magnified tooltip inset near each interactionREADME demos, product pages
Cinematiccinematic_exportCamera crops and pans to follow each interactionSocial media, dramatic reveals
Plainconvert_to_gif / convert_to_mp4Standard screen recording, no zoom effectsBug reports, QA captures

Just Say Where It's Going

You don't need to know viewport sizes or formats. Just tell your AI the destination:

"Record a demo for GitHub README"           → 1280×720 GIF
"Record my app for Instagram Reels"         → 1080×1920 MP4
"Make a TikTok demo of my dashboard"        → 1080×1920 MP4
"Record for YouTube"                        → 1280×720 MP4
PlatformSizeFormatAspect
github / readme1280×720GIF16:9
youtube / twitter1280×720MP416:9
reels / tiktok / shorts1080×1920MP49:16
instagram / linkedin1080×1080MP41:1
CustomAny sizeYour choiceAny

Quick Start

Node.js ≥ 20 and ffmpeg required.

# Add to Claude Code
claude mcp add pagecast -- npx -y @mcpware/pagecast

# Or run directly
npx @mcpware/pagecast

# Headless mode (no visible browser)
claude mcp add pagecast -- npx -y @mcpware/pagecast --headless

# First time: install browser
npx playwright install chromium

How It Works

You: "Record a demo of my app"
            ↓
AI → record_page(url, platform: "github")
    Opens Chromium (visible) at 1280×720
    Injects cursor highlight + click ripple
            ↓
AI → interact_page(click, type, hover...)
    Each action records bounding box + timestamp
            ↓
AI → stop_recording
    Saves .webm + timeline.json
            ↓
AI → smart_export (tooltip mode)
        or cinematic_export (crop-pan mode)
        or convert_to_gif (plain mode)
            ↓
    Shipping-ready .gif or .mp4

What makes the demo look professional:

  • Cursor highlight — red dot tracks the mouse so viewers can follow the action
  • Click ripple — visual feedback on every click
  • Tooltip overlay — magnified close-up appears near each interaction, with a small arrow pointing toward it
  • Cinematic pan — smooth crop transitions between interaction targets with easeInOut curves

MCP Tools

ToolWhat it does
record_pageOpen a URL, start recording. Auto-injects cursor highlight + click ripple
interact_pagescroll, click, hover, type, press keys, select, navigate, waitForSelector — all captured with bounding boxes
stop_recordingStop and save .webm + -timeline.json (event timeline with interaction positions)
smart_exportTooltip overlay — magnified tooltip close-up on each interaction
cinematic_exportCinematic crop-pan — camera follows the action between interaction targets
convert_to_gifWebM → optimized GIF (ffmpeg two-pass palette, configurable FPS/width/trim)
convert_to_mp4WebM → MP4 (H.264, ready for social/sharing/embedding)
record_and_exportAll-in-one: record → auto-export to GIF or MP4 based on platform
list_recordingsList all .webm, .gif, and .mp4 files in output directory

Comparison

AutomatedInteractionsDemo zoomOutputAI-drivenPlatform presets
Pagecast✅ click/type/scroll/hover✅ tooltip + cinematicGIF + WebM + MP4
Screen Studio❌ manual✅ cursor-basedMP4
AutoZoom❌ manual✅ click-basedMP4
Playwright MCPRaw .webmPartial
gifcap.dev / Peek / Kap❌ manualGIF
VHS (Charmbracelet)✅ scriptedTerminal onlyGIF

Screen Studio and AutoZoom have great zoom — but require manual recording. Pagecast is the only tool where the AI records AND the demo effects are automatic.

Configuration

SettingDefaultNotes
BrowserHeaded (visible)--headless for background
GIF FPS12Higher = smoother, larger
GIF width800pxHeight auto-scaled
Tooltip magnify1.6xHow much the tooltip zooms in
Tooltip size380pxSize of the tooltip inset
Cinematic zoom2.5xHow much the camera zooms in
Zoom transition0.35sSmoothstep ease-in/out duration
Cursor overlayOnRed dot + click ripple effect
Video viewport1280×720Or use platform parameter
Output dir./recordingsOverride: RECORDING_OUTPUT_DIR

Architecture

src/
├── index.js        # MCP server — 9 tools, platform presets, stdio transport
├── recorder.js     # Playwright browser lifecycle + sessions + event timeline
├── converter.js    # ffmpeg GIF/MP4 + tooltip overlay + cinematic zoom conversion
├── zoom.js         # Zoom engine — chains, panning, tooltip events, FFmpeg expressions
├── tooltip.js      # Tooltip PNG generator (rounded rect + arrow, pure Node.js)
└── remotion/
    ├── ZoomComposition.jsx  # React composition for cinematic zoom
    ├── Root.jsx             # Remotion entry point
    └── render.js            # Remotion CLI wrapper
  • Event timeline — every interaction records bounding box + timestamp
  • Cursor overlay — red dot + click ripple injected into the page
  • Tooltip overlay — magnified close-up in a clean tooltip frame with directional arrow
  • Cinematic zoom chains — nearby interactions form chains that zoom in, pan, zoom out
  • Thread-safe FFmpeg expressions — crop filters work correctly with multi-threaded encoding
  • Headed by default — watch what the AI does
  • execFile not exec — safe against shell injection

More from @mcpware

ProjectWhat it doesInstall
Instagram MCP23 Instagram Graph API tools — posts, comments, DMs, stories, analyticsnpx @mcpware/instagram-mcp
Claude Code OrganizerVisual dashboard for Claude Code memories, skills, MCP servers, hooksnpx @mcpware/claude-code-organizer
UI AnnotatorHover labels on any web page — AI references elements by namenpx @mcpware/ui-annotator
LogoLoomAI logo design → SVG → full brand kit exportnpx @mcpware/logoloom

License

MIT

Files in the repo

Repository payload19 top-level entries
  • .claude
  • .claude-plugin
  • .github
  • docs
  • src
  • test-recordings
  • .gitignore
  • .npmignore
  • Dockerfile
  • glama.json
  • LICENSE
  • llms.txt
  • package-lock.json
  • package.json
  • README.md
  • README.zh-HK.md
  • record-github-demos.mjs
  • server.json
  • test-demo.mjs

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More connectors

High-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph — average repo in milliseconds. 158 languages, sub-ms queries, 99% fewer tokens. Single static binary, zero dependencies.

43k

Universal provider proxy for OpenAI Codex & Claude Code — use any LLM (Claude, Gemini, Grok, DeepSeek, Ollama…) with Codex CLI, App, SDK, and Claude Code

14k
okf-memory/
okf-agent-memory

Git-native persistent memory for AI coding agents. Implements Google OKF v0.2 with sub-300µs in-memory BM25 search, embedded MCP server, and progressive disclosure. Slashes token bloat by 80% with zero external databases or dependencies. Built in pure Go.

547
tirth8205/
code-review-graph

Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo workflows.

31k
2akouwu/
reverify

Stop your AI from making things up — it proposes, deterministic tools decide, every claim checked against ground truth with evidence. Grounded facts and context survive resets. Reverse engineering is the proving ground. MCP server + CLI.

1.1k
t8y2/dbxConnectors

20 MB lightweight cross-platform database client for 90+ databases, including MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, SQL Server, and Dameng. Built-in AI, MCP Server, CLI, desktop and Docker. | 轻量级跨平台数据库管理工具,支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、达梦等 90+ 数据库,提供桌面端、Docker、CLI、内置 AI 助手和 MCP Server。

19k